2018 IEEE Nuclear Science Symposium and Medical Imaging Conference
To search for a specific ID please enter the hash sign followed by the ID number (e.g. #123).


Session chair: Metzler, Scott, D. (University of Pennsylvania, Department of Radiology, Philadelphia, US); Rafecas, Magdalena, (University of Lübeck, Institute of Medical Engineering, Lübeck, Germany)
Shortcut: M-03
Date: Wednesday, 14 November, 2018, 1:40 PM
Room: Park Side Ballroom
Session type: MIC Session


Click on an contribution to preview the abstract content.

Poster panel: 1

Poster Number:

Development of a depth of interaction detector using a multi-pixel photon counter module for PET (#1119)

K. Hakamata1, H. Uchida1, T. Sakai1, H. Yamauchi1, K. Shimizu1

1 Hamamatsu Photonics K.K., Hamamatsu, Japan


A novel depth-of-interaction (DOI) detector with a single-ended readout was proposed previously. One crystal unit of the detector consisted of a U-shaped transform of a dual-ended readout detector constructed using two scintillator pillars segmented by sub-surface laser engraving (SSLE). The two pillars were coupled with optical glue at the top segment, and an ESR film was inserted between them, running from the second segment to the bottom. Two multi-pixel photon counters (MPPCs) were coupled at the bottom of the pillars, and DOI information was obtained through a centroid calculation of their outputs. We applied this method to a MPPC module (HAMAMATSU C13500-3075LC-16), which has a 16×16 MPPC array (size of one MPPC is 3 mm × 3 mm), and individual time and energy information is acquired by signal processing boards inside the module. Using time and energy information, identification of the crystal unit and DOI segment can be performed. A 6-layer DOI crystal array was coupled to the MPPC module and basic performances were evaluated.

Keywords: PET, DOI-detector
Poster panel: 4

Poster Number:

A Metamaterial approach in the frame of the 10ps Time-of-Flight PET challenge (#1291)

P. Lecoq1, G. Bizzari2, R. Martinez Turtos1, E. Auffray Hillemanns1, S. Gundacker1

1 CERN, Geneva, Switzerland
2 Cranfield University, Cranfield, United Kingdom


The 10ps TOFPET challenge requires a parallel effort on all the components of the detector chain (crystal, photodetector, readout electronics). We have demonstrated that even with the best presently available SiPM photodetectors, the most popular scintillators for PET (L(Y)SO, L(G)SO, LaBr3) are limited in their timing performance at low energy and that a mechanism to produce a few hundred prompt photons on top of the scintillation signal is necessary in order to reach 10ps coincidence time resolution (CTR) at 511keV.

This contribution describes our work for developing a metamaterial-based scintillator block having these characteristics. A quantitative justification will be given, based on our measurements, for a combination of L(Y)SO, in form of block, plates or fibers with:

  • ZnO rods of different sizes
  • CdSe nanoplatelets, 35x8x1nm3, produced by wet technology, taking advantage of the 1D quantum confinement, allowing coherent multi-excitonic emission with a <1ps rise time and 290ps decay time,
  • ZnO:Ga quantum dots, produced by photo-induced precipitation with calcination in air and Ar/H2 atmosphere, with <1ps rise time and 504ps decay time,
  • CsPbBr3 perovskite quantum dots showing a bright multiexcitonic emission with rise time of <1ps, and a decay time of 320ps (»50%) and 2.48ns (»50%).


Simulation results for different, but not yet optimized, geometrical configurations of the metamaterial structure show that an energy leakage from LYSO to the fast component of the metastructure of ≥ 100keV can be obtained for about 20% of the 511keV g-rays fully converted in the metastructure, inducing the production of »250 prompt photons on CdSe nanoplatelets on top of the LYSO scintillation signal.

Evidence of this prompt emission has been obtained in our lab on a simple heterostructure made of 200micron thick LYSO plates covered by CdSe nanoplatelets layers deposited by drop casting. Results on more metastructures configurations will be given at the time of the conference.

Poster panel: 7

Poster Number:

Reproduction of response functions of a multi-pixel-type energy-resolved photon counting detector while taking into consideration interaction of X-rays, charge sharing and energy resolution (#1327)

N. Kimoto1, H. Hayashi1, T. Asahara2, E. Tomita1, S. Goto1, Y. Kanazawa2, S. Yamamoto3, M. Okada3, M. Yamasaki3

1 Kanazawa University, Division of Health Sciences, Graduate School of Medical Sciences, Kanazawa, Japan
2 Tokushima University, Graduate School of Health Sciences, Tokushima, Japan
3 Job Corporation, Yokohama, Japan


Energy-resolved photon counting detectors (PCD) are currently being developed for medical application. It is hoped that these detectors can be used for deriving precise material information. Many researches are concerned with the necessity of considering the response of PCD for analysis of measured spectra. We plan to apply a correction for incomplete energy signals using software applications. For establishing the correction procedure, we should know the response of PCD in terms of interactions between detector materials (Cd, Zn and Te), charge sharing effect, and energy resolution. First, to derive an ideal response of PCD “R1”, Monte-Carlo simulation is carried out. In the simulated R1, characteristic X-ray peaks of Cd and Te were clearly observed; these peaks were produced in the multi-pixel-type detector. Second, taking into consideration charge sharing effect and energy resolution, response function “R2” was determined; constant background-type charge sharing function and energy dependent Gaussian function were assumed based on published researches. Comparing the experimental spectra (50 kV & 80 kV) measured with our test-model detector, parameters for R2 were optimized. As a result, we can reproduce X-ray spectra measured with multi-pixel-type PCD; typical parameters for R2 are peak efficiency 25% and energy resolution 8% at 80 keV. Next, using the X-ray spectra folded with R1×R2, ratios of full-energy peaks in the spectra were analyzed. X-ray attenuation of aluminum having a thickness of 1 cm was calculated to mimic dental radiography. In our application for material identification, attenuation coefficient mt should be determined from the measured spectra. When tube voltage of 80 kV was applied, obtained mt for 50-80 keV is in good agreement with the theoretical values. Based on the present research, the knowledge of response function will be applied to our material identification method which was proposed using ideal X-ray spectra.

Poster panel: 10

Poster Number:

A novel DOI Estimation Approach for Monolithic Scintillator Crystals in PET Calibratable with Both Experiment and Simulation Based on the Supervised Machine Learning Technique Gradient Tree Boosting (#1429)

F. Müller1, R. Hetzel1, D. Schug1, J. Grahe1, P. Hallen1, V. Schulz1

1 RWTH Aachen, Physics of Molecular Imaging Systems (PMI), Aachen, North Rhine-Westphalia, Germany


Monolithic crystals are considered as an alternative for segmented scintillator arrays in positron emission tomography (PET). Monoliths provide good spatial, timing and energy resolutions as well as intrinsic depth of interaction (DOI) encoding. DOI allows reducing parallax errors (radial astigmatism) at off-center positions within a PET ring. This is especially important for high-resolution PET systems such as organ-specific applications (e.g., breast or neuro-imaging) and total body PET. We present a novel DOI estimation approach which can be calibrated with both experimental and optically simulated data. The DOI estimation is based on the supervised machine learning algorithm Gradient Tree Boosting (GTB). GTB builds a predictive regression model based on a set of sequential binary comparisons (decision trees). GTB models have been shown to be implementable in FPGA if the memory requirements fit the available resources. We propose two optimization scenarios for the best possible DOI performance: One restricting the available memory to enable a future FPGA implementation and one without any restrictions. The DOI performance of the GTB models is compared with a DOI estimation method based on an isotonic regression (IR) of a DOI observable. The DOI observable is defined as the ratio of the highest photon count to the sum of the photon counts of adjacent sensor channels. The usage of a DOI observable is comparable to other methods presented in the literature. Among others, we calculate the spatial resolution (SR) as the FWHM, the bias vector and the 90th percentile distance of the positioning error distribution to quantify the DOI performance. For a 32 mm x 32 mm x 12 mm large monolith, we achieve an averaged SR of 2.16 mm FWHM and 1.98 mm FWHM for the IR and GTB using experimental data, respectively. In contrast to the IR models, the GTB models show a nearly uniform DOI performance over the whole crystal depth.

Keywords: PET, Monolithic Crystal, DOI, Depth of Interaction, Machine Learning
Poster panel: 13

Poster Number:

Performance Comparison of Two Dual-Ended Readout PET Detectors Using Different SiPM Arrays (#1523)

Q. Yang1, Z. Kuang1, X. Fu1, B. Zhao1, Y. Yang1, J. Du1

1 Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, SHENZHEN, China


Two depth encoding PET detectors using dual-ended readout with latest SiPMs from Hamamatsu and Broadcom were evaluated as the candidates for an MRI-compatible brain PET insert which is under development at our lab. The two types of SiPM arrays are both 8 × 8 array of SiPMs. The first one is the Hamamatsu S14161-4050HS-08 monolithic SiPM array, and the second one is fabricated by tiling four Broadcom 4 x 4 monolithic AFBR-S4N44P163s in a 2 x 2 configuration. The Hamamatsu SiPM arrays have a surface area of 33.8 × 33.8 mm2 and a pitch size of 4.2 mm, and the Broadcom SiPM arrays have a surface area of 32.02 × 32.02 mm2 and a pitch size of 3.93 mm. The active pixel areas of the Hamamatsu SiPMs and the Broadcom SiPMs are 4.0 × 4.0 mm2 and 3.72 × 3.72 mm2 respectively. Each detector module consists of two identical SiPM arrays coupled to both ends of a 20 × 20 array of 1.42 × 1.42 × 20 mm3 polished LYSO crystals (80 um BaSO4 reflector). In this paper, the performance of the two detector modules were measured and compared at different bias voltages and a temperature of 22.8 °C ± 0.3 °C. The flood histograms show that all the LYSO elements of the two detector modules can be clearly identified. The best flood histogram quality parameter is 4.3 ± 0.9 (obtained at a bias voltage of 41.5 V) for the detector based on the Hamamatsu SiPMs, and 3.4 ± 0.9 for the detector based on the Broadcom SiPMs (obtained at a bias voltage of 30.5 V). At those bias voltages, the energy resolution, DOI resolution and timing resolution are 10.5 ± 3.4 %, 1.88 ± 0.16 mm and 861 ± 246 ps for the detector based on the Hamamatsu SiPMs, and 13.1 ± 5.6 %, 1.90 ± 0.14 mm and 873 ± 201 ps for the detector based on the Broadcom SiPMs. Both detectors are good candidates for our on-going brain PET insert and detector based on Hamamatsu SiPMs provides a better overall performance.

Keywords: Brain PET; PET detector; SiPM; Depth-of-Interaction
Poster panel: 16

Poster Number:

First experimental results for novel gamma detector for multi-pinhole molecular breast tomosynthesis (#1578)

B. Wang1, R. Kreuger1, J. Huizenga1, F. J. Beekman1, 2, M. C. Goorden1

1 Delft University of Technology, Department Radiation Science and Technology, Section Biomedical Imaging, Delft, Netherlands
2 MILabs B.V., Utrecht, Netherlands


PMT-based scintillation gamma detectors are still predominant in medical imaging due to their cost-effectiveness. A drawback of these detectors is that the position estimation is severely degraded near the edges of the scintillator, which is not acceptable for the multi-pinhole molecular breast tomosynthesis scanner we are currently developing. Here we show first experimental results of a novel scintillation detector geometry with an unconventional light-guide-PMT layout that aims to solve the dead-edge issue. To this end monolithic NaI(Tl) scintillator is read out by 15 square PMTs. Instead of attaching a full row of PMTs near the scintillator’s edge, we inserted additional light-guides at the upper (critical) edge, imitating overextended PMTs without requiring additional space at the critical edge that is placed near the woman’s chest wall. The detector is scanned in a 5 mm grid by a collimated Tc-99m beam to obtain the reference PMT outputs that are used as input of a maximum likelihood (ML) positioning algorithm. Another scan of the collimated beam is done in a 10 mm grid to evaluate the spatial resolution and positioning linearity over the detector. From the measurements, we find that the position estimation with the ML algorithm is linear over the detector area. The average spatial resolution in horizontal and vertical directions are 3.4 and 3.6 mm respectively. The dead edge is smaller than 3 mm. To conclude, this gamma detector is a good candidate for multi-pinhole molecular breast tomosynthesis which requires small dead edges.

Keywords: gamma detector, Anger camera
Poster panel: 19

Poster Number:

Impact on the time resolution of double sided crystal readout in TOF-PET (#1605)

S. Gundacker1, 2, E. Auffray Hillemanns1, P. Lecoq1, M. Paganoni2

1 CERN, Geneva, Switzerland
2 UniMIB, Milano, Italy


A precise time of flight (TOF) estimation of the coincident 511keV anihilation gammas in positron emission tomography (PET) can give a more than tenfold improvement in image signal to noise ratio. Fast scintillating crystals coupled to compact silicon photomultipliers (SiPMs) constitute a versatile system for PET with state-of-the-art coincidence time resolutions (CTRs) ∼80ps FWHM for the best configurations, using small LSO:Ce:Ca scintillators of 2x2x3mm3 size. How to retain this time resolution or achieve a CTR of better than 100ps FWHM with crystals used in PET of typically 20mm length is still an open question. In this work the timing performance of LYSO:Ce and LSO:Ce codoped Ca scintillators of different lengths, wrappings, surface finishing and cross-sections read out with SiPMs on both crystal extremities are presented. We demonstrate that applying double sided readout and 511keV gamma depth-of-interaction (DOI) time correction a CTR of 105±4ps FWHM can be achieved using a fully polished 2x2x20mm3 LSO:Ce codoped with 0.4%Ca crystal coupled to HPK S13360-075P SiPMs. For this particular setup the DOI can be reconstructed within 3.8mm precision applying standard light amplitude measurements and within 6.7mm using only delay time measurements. Even longer LYSO:Ce bars of 3x3x50mm3 can reach CTR values of 140±5ps FWHM, along the whole bar, if DOI corrrections are applied. Such results are interesting for alternative PET designs or in high energy physics detectors. It can be concluded that double sided readout shows a promising way to achieve 100ps in whole body PET with available technologies and will be further discussed in the frame of the 10ps TOF-PET challenge, where DOI corrections will become indispensable.


Keywords: 100ps TOF-PET, depth of interaction, double sided readout
Poster panel: 22

Poster Number:

Measurements with the PETsys TOFPET2 ASIC Evaluation Kit using KETEK and FBK SiPM-crystal configurations (#1719)

V. Nadig1, 2, D. Schug1, 2, B. Weissler1, P. Gebhardt1, 2, V. Schulz1, 2

1 RWTH Aachen University, Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, Aachen, North Rhine-Westphalia, Germany
2 IEEE Member, New York City, United States of America


A precise measurement of the time-of-flight information in positron emission tomography in the order of few hundred pico-seconds is key to the applicability of this method in clinical practice. In prior studies, a new application-specific integrated circuit developed by PETsys Electronics SA showed promising results regarding the time and energy resolution in prior studies with analog silicon-photomultipliers. In this work, we present the performance of the circuit with different analog silicon-photomultiplier-crystal configurations for different temperatures, discriminator thresholds and overvoltages. All employed configurations reached a time resolution in the sub-ns range, while the best results were achieved by a FBK (NUV-HD) silicon-photomultiplier with a 2.62 mm x 2.62 mm x 3 mm Lutetium-yttrium oxyorthosilicate crystal reaching a coincidence resolution time of 162.7 ± 2.6 ps (FWHM) at 6.25 V overvoltage and a discriminator threshold setting of 30. Both, the mentioned FBK (NUV-HD) configuration and the KETEK-3315-WB silicon-photomultiplier with the same crystal reached stable energy resolution under 10 %. Providing options of of valid configurations this study forms the basis for further investigations regarding the applicability of the application-specific circuit in a simultaneous positron-emission-tomography/magnetic-resonance scanner.

Keywords: PET, time-of-flight, analog SiPM, data acquisition circuits
Poster panel: 25

Poster Number:

Experimental studies of the DOI decoding accuracy of PET detectors with light sharing window (#1911)

X. Zhang1, S. Xie1, J. Yang1, H. Peng1, Q. Huang2, J. Xu1, Q. Peng3

1 Huazhong University of Science and Technology, Wuhan, China
2 Shanghai Jiao Tong University, Shanghai, China
3 Lawrence Berkeley National Laboratory, Berkeley, California, United States of America


Positron emission tomography system without depth of interaction decoding capability has parallax error problem deteriorating the image quality in peripheral FOV. Light-sharing-window, which means replacing the reflection between segmented crystals by optical glue physically, can guide the photons to the surrounding crystal and encode the DOI information. The purposes of this study is to investigate LSW and the decoding method for DOI measurement by experiment. In this work, we developed a detector with LSW. By analyzing the flood map and conducting parallel exposure experiment, we defined the relationship between map pixel and continuous DOI information. (1) 6 mm SiPMs has better decoding performance than 3 mm SiPMs. (2) The best MAE resolution is 0.8mm in central area and the average resolution is 1.182 mm. (3) We can optimize window height and SiPM coupling method to achieve higher DOI resolution. (4) The average MAE resolution improves when the segmented crystal numbers increase.

Light sharing window method is a very promising method for accurate DOI measurements. Our next step is to conduct simulation to find out appropriate window settings with LSW. Experiments of different window height samples and two coupling methods would be carried out.

Keywords: light sharing window, PET detectors, DOI decoding, LYSO crystal
Poster panel: 28

Poster Number:

Simulation Studies for an Optical Property Modulation-Based Radiation Detection Concept for PET (#1967)

L. Tao1, J. Wang1, M. Esmaeelpour1, C. S. Levin1

1 Stanford University, Radiology, Stanford, California, United States of America

This work is partly supported by NIH grant 5R01EB02390302.


The concept of using modulation mechanisms of material’s optical properties for annihilation photon detection has been proposed by our group as a potential method to significantly improve the coincidence time resolution of PET detectors. In this paper, we use simulations to study whether the optical property modulation effect induced by single 511 keV photon interactions is detectable using modern optics methods. We start by simulating (with pyPENELOPE) the energy deposition profile of an ejected electron in the detector crystal, and convert it to the number of ionization charges created. Then we simulate (with Lumerical) the distribution of ionization charge induced modulation in carrier density and electric field. Eventually we calculate the change in optical photo-absorption and photo-refraction caused by the modulation in charge density and local electric field strength resulting from ionization. We focus our simulation work on cadmium telluride (CdTe) and indium antimonide (InSb) crystals under zero bias voltage at room temperature. Based on our calculation and simulation results, the largest optical photo-absorption modulation induced by a 511 keV photon is around 1 cm-1(in CdTe), which indicates a total probe laser intensity change of 1.5%. The largest photo-refraction modulation is around 5×10-5 (in InSb), which represents a phase change of 6 milliradians (mrad). With optics experiments, absorption measurement sensitivity has been reported to be able to detect a total probe laser intensity change of 0.0007%. Refraction measurement sensitivity has been reported as better than 1mrad. Therefore our simulation results show that with appropriate experiment design, it is possible to detect individual 511 keV photons with the proposed optical property modulation-based detection method. 

Keywords: ToF-PET, optical photo-absorption modulation, photo-refraction modulation, CdTe, InSb
Poster panel: 31

Poster Number:

An improved photon collimation technique for monolithic PET detector calibration (#2000)

A. González-Montoro1, 2, L. A. Pierce2, W. C. J. Hunter2, P. E. Kinahan2, A. J. González1, R. S. Miyaoka2

1 Universitat Politècnica de València, Institute for Instrumentation in Molecular Image (i3M) -CSIC, Valencia, Spain
2 University of Washington, Department of Radiology, Seattle, Washington, United States of America


The calibration of monolithic positron emission tomography (PET) detector modules often uses data from a collimated beam of 511 keV photons that is scanned across the detector face in order to characterize the light output from the detector. The standard method for forming the calibration beam typically uses a small gamma source and a small coincidence crystal to perform coincidence collimation. This “hardware collimation” technique requires that the coincidence crystal be carefully aligned with the source and that the source and crystal be moved synchronously during data collection. This necessitates a complicated hardware setup that cannot, in general, be used within an assembled PET scanner. For this reason, recalibration of a monolithic PET detector using hardware collimation would require removal of the module from the scanner, which can be a sensitive and time-consuming process.


The method of “software collimation” places a positron source in close proximity to the entrance face of the detector module and uses software to filter the data to simulate a collimated calibration beam. This method eliminates coincidence crystal/point source misalignment issues, requires less hardware, and allows for the collection of calibration data while a detector is still mounted within a PET scanner. Multi-source software collimation has the potential to speed the calibration times significantly.


Despite the advantages of software collimation, it has not been widely adopted. In this work, we perform hardware collimation and software collimation on the same monolithic PET detector module, and the resulting calibration datasets are compared.


Our results indicate that software collimation achieves the same spatial resolution as hardware collimation while using less hardware and requiring no beam alignment. Moreover, since it is possible to define the acceptance angle, it is possible to improve the source/beam positioning accuracy and therefore the system calibration performance.

Poster panel: 34

Poster Number:

Development of the 4-Layer GAGG DOI Detector for a Scatterer of Compton Camera (#2086)

S. Takyu1, E. Yoshida1, N. Inadama1, F. Nishikido1, M. Nitta1, K. Kamada2, K. Parodi3, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology (QST), National Institute of Radiological Sciences (NIRS), Chiba, Japan
2 C&A corporation, Sendai, Japan
3 Ludwig-Maximilians Universität München, Munich, Germany


We have developed the first prototype based on our new concept of whole gamma imaging (WGI) which is utilizing all detectable gamma rays for imaging by combining the two technologies of PET and Compton imaging. In the prototype we developed in the last year, the scatter detector ring consisted of 40 single-layer detectors, each of which had a 24 x 24 array of Gd3Al2Ga3O12(Ce) (GAGG) crystals 0.9 x 0.9 x 6 mm3 in size with a multi-pixel photon counter (MPPC) array. However, as the 6 mm GAGG thickness was originally chosen for gamma rays around 511 keV, the sensitivity of Compton imaging for gamma rays with higher energy such as 1157 keV was not good enough. In this study, therefore, we developed a four-layered depth of interaction (DOI) scatter detector for our next advancement. On the basis of the light sharing method we developed for DOI capability, a 14 x 14 x 4-layered GAGG crystal block, which is composed of GAGG crystals with 1.45 x 1.45 x 4.5 mm3 in size, was placed on the MPPC array module with the resistor network (MPPC: 3 x 3 mm2, 8 x 8 array, pixel size of 50 x 50 μm2). In crystals in the central part of the position map obtained by the Anger-type calculation, averaged pulse height and energy resolution at four gamma ray energies (202, 307, 511 and 662 keV) was investigated in each layer. In the central area, all of the crystals in four layers could be clearly identified. The DOI detector showed better pulse linearity than the single-layer detector we developed in the last year. This implies the effect of MPPC saturation for the DOI detector was smaller than that for the single-layer detector. It was presumed that scintillation light spread by using the light sharing method for the DOI capability reached the MPPC surface. Energy resolution of the DOI detector in all layers was comparable with that of single-layer detector. In conclusion, we succeeded in developing the four-layer GAGG DOI detector with high sensitivity for high energy gamma rays.

Keywords: Scatterer, Compton imaging, Four layer DOI
Poster panel: 37

Poster Number:

A High Count-Rate and Depth-of-Interaction Resolving MR-Compatible Single Layered One-Side Readout Pixelated Scintillator Crystal Array for PET Applications (#2207)

J. M. C. Brown1, 2, D. R. Schaart1

1 Delft University of Technology, Radiation Science & Technology, Delft, Netherlands
2 University of Wollongong, Centre for Medical Radiation Physics, Wollongong, Australia


Organ-specific, targeted Field of View (FoV) Positron Emission Tomography (PET)/ Magnetic Resonance Imaging (MRI) inserts are viable solutions for a number of imaging tasks where whole-body PET/MRI systems lack the necessary sensitivity and resolution. Whilst these systems' smaller PET bore diameters result in increased sensitivity across their FoV, it also increases the impact of ``parallax error'' on system spatial resolution that arise from Depth-of-Interaction (DoI) effects within clinical PET radiation detectors. To suppress these effects, and reach the target spatial resolutions of 1 mm, compact SiPM-coupled crystal arrays with adequate x-y and DoI resolution are required without compromising energy resolution, timing resolution and maximum count rate. This work outlines the development of a novel MR-compatible high count-rate and DoI resolving two-axis patterned reflector strip pixelated scintillator crystal array design for organ-specific, targeted FoV PET/MR inserts. A controllable light-sharing approach was develop through a repeating phased open reflector cross-section pattern along each light-sharing axis. This novel design creates virtual light trapping boundaries and enables the determination of DoI whilst minimising the probability of event pile-up. Here, we present the design philosophy behind these PET detectors, in-silico optimisation of a version intended for the HYPMED PET/MR insert, and discussion on the results of an experimental prototype under construction at TUDelft.

Poster panel: 40

Poster Number:

Evaluation of Timepix3 based CdTe photon counting detector for fully spectroscopic small animal SPECT imaging (#2255)

E. Trojanova1, J. Jakubek1, D. Turecek1, L. Sefc2, V. Sykora2, P. Francova2

1 ADVACAM s.r.o., Prague, Czech Republic
2 Center for Advanced Preclinical Imaging (CAPI), First Faculty of Medicine Charles University, Prague, Czech Republic


The imaging method of SPECT (Single Photon Emission Computed Tomography) is used in nuclear medicine for diagnostics of various diseases or malfunctions of organs.  The distribution of medically injected, inhaled, or ingested radionuclides in the patient body is imaged using a gamma-ray sensitive camera with a suitable collimator. Combining many such images taken at different observation angles the 3D image is calculated. Most of SPECT systems use scintillator based cameras. A scintillating screen converts gamma radiation to visible light which is collected and converted to electrical signal by the photodetector. This detection principle does not provide good energy resolution and does not allow efficient suppression of unwanted signals such as Compton scattering.

The particle counting Timepix3 based camera with semiconductor sensor has many advantageous properties. It does not suffer from a dead time since it uses event-based readout, it records energy and detection time for each detected ionizing particle, it provides high detection efficiency and high spatial resolution.

The main goal of this work is an evaluation of properties of Timepix3 CdTe detector for SPECT method. The Timepix3 properties such as energy and spatial resolution are exploited for image quality improvement suppressing unwanted signals such as Compton scattering in the sample volume. Thanks to suppression of background is possible to distinguish two neighborhood objects that are not visible using standard scintillator cameras due to the high intensity of the background.

Preliminary measurements were performed on specially prepared plastic phantom filled by radioisotopes and then repeated with the mouse in vivo. Plastic phantom was also used as a sample for the follow-up tomographic scan.

Small size and compact read-out electronics with the excellent connectivity of USB3.0 allows construction of portable system composed of several Timepix3 units suitable for fully 3D SPECT imaging for preclinical and clinical practice.

Keywords: Timepix3, SPECT, energy resolution, spatial resolution, Compton scattering
Poster panel: 43

Poster Number:

Dual mode TOF detectors of Prompt Gamma and Positron Annihilation Gamma for in-beam range verification in pencil-beam Proton Therapy  (#2353)

M. V. Nemallapudi1, S. - C. Lee1, C. - H. Lin1, M. - L. Chu1, A. Chen2, A. Urfa2, S. Ali2, A. Rahman2

1 Academia Sinica, Institute of Physics, Taipei, Taiwan
2 University, National Central University, Taoyuan, Taiwan


The precise dose delivery achieved through Proton therapy makes it an appropriate choice for cancer treatment. To reduce the risk of possible damage to sensitive organs close to the Bragg Peak it is essential to accurately perform range verification. Existing solutions are limited in terms of their timing capabilities and the possibility of realtime verification, especially for Cyclotron beams with a nanosecond level beamstructure. We propose a dual mode detection system based on time of flight detectors for verifying the range by detecting 511 keV gammas coincidences and to obtain additional molecular information through prompt gamma detection. Coincidence detection in our proposed system is performed using LYSO scintillators coupled to silicon photomultipliers (SiPMs) while LaBr crystals of small crossection will be used for prompt gamma detection. We measured coincidence time resolution (CTR) values of around 200 ps FWHM using single crystals of LYSO and LFS (3x3x20mm3) measured with a Na22 source in the laboratory. A 32 channel detection system was used to test the secondaries emitted from a PMMA phantom due to nuclear interactions from a 130 MeV proton beam at the Chang Gung Memorial Hospital (CGMH, Linkou). We observed an exponential decay of 511 keV counts during the beamOFF, and the coincidences from 511 keV photons with a resolution of around 800 ps FWMH over the entire system without energy filtering. This value can be improved with optimization. We measured a variation in counts with depth that closely follows the simulation. We are currently working on an optimized DAQ system to deal with the high rate environment at the proton beam.  Once fully realized, the system will be able to predict accurately the proton range at a millimetre level precision and provide additional information on the molecular composition that can significantly improve the treatment planning while a high retaining patient throughput and causing minimum inconvenience to the patient.

Keywords: SiPMs, crystals, time of flight, in-beam, proton therapy
Poster panel: 46

Poster Number:

Sparse SiPM Pixel Arrangement for Side Readout of Long and Narrow Scintillation Crystal Elements (#2424)

J. W. Cates1, C. S. Levin1, 2

1 Stanford University, Department of Radiology, Stanford, California, United States of America
2 Stanford University, Department of Bioengineering, Stanford, California, United States of America


Positron emission tomography (PET) detectors for clinical systems employ long and narrow scintillation crystal elements optically coupled on their narrow end to a photosensor. The aspect ratio of this traditional configuration yields low scintillation light collection efficiency (LCE) and significant, interaction depth-dependent scintillation light transit time jitter from the 511 keV photon interaction point to the photodetector. Alternatively, coupling the photosensor on the long side of crystal elements results in near-complete (LCE) and low scintillation photon transit time jitter. In this scintillation light readout configuration, the achievable coincidence time resolution (CTR) with 3x3x20 mm3 crystals can achieve ~100 ps FWHM, roughly 40 to 100 ps lower than possible with end readout. This readout technique also intrinsically provides three-dimensional positioning of one or more interactions per event, with sub-mm depth-of-interaction resolution. Since arranging SiPMs along the long side of crystal elements requires a moderate increase in photosensor area, which is associated with a higher cost, it is of interest to study whether a reduced, sparse arrangement of those sensors is possible without compromising performance. In this work, an experimentally validated simulation model was used to investigate various contiguous and non-contiguous sparse arrangements of SiPM pixels and reflectors along the side of long and narrow scintillation crystal elements. Twenty-seven configurations were tested, where the aim was to minimize the photosensor area while also minimizing influence on CTR, depth of interaction resolution, LCE, and energy resolution. An optimum SiPM/reflector configuration was identified, wherein only 60% of the crystal’s side is coupled to SiPMs while maintaining >80% light collection efficiency, depth-of-interaction resolution of ~1 mm, achieving CTR between 100-110 ps FWHM, and energy resolution of 11% FWHM.

Keywords: time-of-flight positron emission tomography
Poster panel: 49

Poster Number:

Tradeoffs in Cherenkov Detection for Positron Emission Tomography (#2511)

A. A. Muntean1, F. Gramuglia1, E. Venialgo2, C. Bruschini1, E. Charbon1

1 EPFL, AQUA STI IMT EPFL, Neuchatel, Neuchâtel, Switzerland
2 TU Delft, AQUA, Delft, Netherlands


The coincidence time resolution (CTR) of a time-of-flight positron emission tomography (TOF-PET) scanner is an essential parameter, which determines signal-to-noise ratio (SNR) in image reconstruction algorithms. CTR has an important impact on scintillation-based detectors used in TOF-PET. Cherenkov PET has emerged as a prompt-photon based radiation detector, which has more accurate light-photon emission statistics. However, light transport becomes a dominant degradation effect in the light-photon detection statistics, which is highly correlated to the crystal length and consequently to the gamma-photon detection efficiency.
This work focuses on finding the optimal design in terms of timing resolution and gamma-photon detection efficiency for Cherenkov detectors. Radiator design considerations such as size, materials and coatings, along with electronic jitter, and dark count rate (DCR), are included in a mathematical simulation that determines their effect on timing resolution. Currently, two different scintillators were simulated, namely PbF2 and PbWO4, with a maximum Cherenkov count per frame of 12 photons; an electronic jitter of 4.3ps to 42.5ps (standard deviation) was added. Simulation results confirm the high impact of jitter on the uncertainty of arrival times of Cherenkov photons.

Keywords: Cherenkov, Time resolution, gamma detection efficiency, electronics jitter, TOF-PET
Poster panel: 52

Poster Number:

Experimental Investigation of Cherenkov Cone Detection for High-Energy Gamma-Ray Imaging (#2614)

R. Bayerlein1, A. B. Brill3, I. Fleck1, L. R. Furenlid4, 5, W. Khalid1, T. E. Peterson2, 3, A. H. Walenta3, 1

1 University of Siegen, Department of Physics, Siegen, North Rhine-Westphalia, Germany
2 Vanderbilt University Medical Center, Institute of Imaging Science, Nashville, Tennessee, United States of America
3 Vanderbilt University Medical Center, Radiology and Radiological Sciences, Nashville, Tennessee, United States of America
4 University of Arizona, College of Optical Sciences, Tucson, Arizona, United States of America
5 University of Arizona, Department of Medical Imaging, Tucson, Arizona, United States of America


In previous simulations we have shown that it may be possible to reconstruct the interaction position of high-energy gamma rays (E>1MeV) in three dimensions based on detection of the two-dimensional distribution of Cherenkov photons on the exit surface of an appropriate radiator material. The ability to do so depends on the multiple scattering of the energetic Compton electron that induces the emission of Cherenkov radiation and the efficient detection of the Cherenkov photons with appropriate spatial sensitivity. Such detection capability would facilitate imaging of the distribution of certain targeted alpha-emitter therapies and also could be applicable in range verification for proton therapy.  For initial proof of concept measurements, the Cherenkov photons created in PMMA by the energetic electrons from a collimated strontium source have been detected in coincidence using a 4x4 silicon photomultiplier (SiPM) array. The data exhibit the expected variation with source position, and the average number of detected photons is consistent with expectations based on simulations and the SiPM efficiency. Further measurements of photon yield as a function of energy and their spatial distribution will provide the information necessary to estimate the expected efficiency and spatial resolution attainable in a larger-scale detection system.

Keywords: Cherenkov radiation, Gamma-ray imaging
Poster panel: 55

Poster Number:

Comparison of Different Readout Methods of SiPM Array Coupled to LYSO Array for the Next Generation EXPLORER PET Scanner (#2717)

E. Mikhaylova1, J. Du1, X. Bai1, M. S. Judenhofer1, S. R. Cherry1

1 University of California, Davis, Biomedical Engineering, Davis, California, United States of America


The aim of this work is to optimize the performance of the current EXPLORER PET detector for possible use in the next generation total-body scanner. For this purpose, we designed and compared 4 different configurations of multi-resolution SiPM arrays coupled to the same LYSO array as the one used in the EXPLORER PET system. The SiPM array consist of 13 SensL J-Series SiPMs of different size, 6 × 6 mm2 and 3 × 3 mm2, and covers a 20 × 20 mm2 area. Each SiPM is read out individually.  The dimensions of the LYSO array are 17 × 20 × 18 mm3. It’s made of 7 × 6 crystals with dimensions of 2.8 × 2.8 × 18 mm3. We obtained flood histograms, energy spectra, light yield and energy resolution of the LYSO array using four different SiPM readout methods:  A: SiPMs in the array; B: 5 SiPMs in the array; C: 9 SiPMs in the array and D: all 13 SiPMs in the array. Coincidence data were acquired at room temperature at 29.0 V SiPM bias voltage. Results: All methods are able to identify all 42 LYSO crystals. Method D gives the best flood histogram while Method A shows the poorest. Method D also shows the highest light yield. The energy resolution results (entire block) are: A. 23.7% B. 15.8% C. 21.7% D. 21.1%. The best computed timing resolution for the whole LYSO array is 588 ps. Conclusions. A flexible multi-resolution detector module has been designed and developed to enable the trade-offs in SiPM coverage and sampling of a scintillator array suitable for clinical PET applications to be studied. Initial results have been obtained and individual crystal-based energy and timing resolution measurements are expected to further improve the results.

Keywords: EXPLORER PET, SiPM readout optimization, LYSO array
Poster panel: 58

Poster Number:

Exploring Capability of Decoding DOI using Waveforms from Dual-ended Readouts in a Convolutional Neural Network (#2918)

C. - C. Liu1, E. Roncali1, S. I. Kwon1, A. R. Selfridge1, S. R. Cherry1, J. Qi1, M. S. Judenhofer1

1 University of California, Davis, Department of Biomedical Engineering, Davis, California, United States of America


The spatial resolution of positron emission tomography (PET) can be improved at the edge of the field-of-view (FOV), if the depth-of-interaction (DOI) information is available in the image reconstruction. The objective of this study was to estimate DOI using a convolutional neural network (CNN) method and to compare the results with the conventional DOI ratio method. GATE V8.0 was used to simulate optical photons propagating along a single LSO crystal with Teflon reflector. The surface treatment of the crystal was polished. The simulated optical photons were collected by two SiPMs placed at both ends of the crystal to obtain signal waveforms. The DOI was predicted by a CNN model after training with 1100 pairs of waveforms at each depth (1-mm step size) across the entire crystal. With the test data, including eight depths not in the training data, the peak positioning errors of the conventional DOI ratio and CNN methods are 0.57mmand 0.38 mm,respectively. The FWHM for the conventional DOI ratio and CNN methods are 5.53 mm and 3.77 mm, respectively. The mean absolute errors (MAEs) of the DOI ratio and CNN methods are 1.92 mm and 1.32 mm, respectively. The CNN method showed better DOI decoding performance than the conventional DOI ratio method. We will investigate the CNN method with more training data and more crystal configurations.

Keywords: PET, Depth-of-interaction, machine learning, convolutional neural network, GATE V8.0
Poster panel: 61

Poster Number:

MTF simulation of a-Se x-ray detectors using ARTEMIS for breast imaging applications (#1125)

Y. - H. Cheng1, 2, J. - A. Lin3, Y. Fang4

1 University of Michigan, Department of Biomedical Engineering, Ann Arbor, Michigan, United States of America
2 Food and Drug Administration, Division of Imaging, Diagnostics, and Software Reliability, Office of Science and Engineering Laboratories, Center for Devices and Radiological Health, Silver Spring, Maryland, United States of America
3 Food and Drug Administration, Division of Biostatistics, Office of Surveillance and Biometrics, Center for Devices and Radiological Health, Silver Spring, Maryland, United States of America
4 Food and Drug Administration, Diagnostic X-ray Systems Branch, Division of Radiological Health, Office of In Vitro Diagnostics and Radiological Health, Center for Devices and Radiological Health, Silver Spring, United States of America


Radiological imaging of breast anatomy for clinical diagnostic and screening purposes requires that x-ray systems have high spatial resolution in order to detect microcalcifications and small lesions. This imaging requirement has driven the development of high spatial resolution imagers with small pixel pitch. In this work, we present a study of the resolution performance of amorphous selenium (a-Se) using the ARTEMIS Monte Carlo code for simulation of direct-conversion x-ray detectors. The model includes photoelectric absorption, Compton scattering, Rayleigh scattering and fluorescent x-ray generation for x-ray interactions.  Also included are high-energy electron scattering interactions that deposit kinetic energy; and electron-hole pair interactions taking into account recombination in the presence of neighboring carriers including Coulomb field and trapping in the presence of external applied electric field. The detector modulation transfer function (MTF) was simulated for mono-energetic x-rays of 12, 13 and 25 keV photons in the x-ray energy range commonly used in breast imaging, and pre-sampled MTF was simulated using RQA-M2 to model a realistic clinical x-ray beam. The simulated pre-sampled MTF was compared with experimental measurements from a commercially available a-Se full-field digital mammography system and showed a percentage difference of 3%. The ARTEMIS simulation model can be used to study the MTF performance of a-Se solid state x-ray detectors for breast imaging applications.

Keywords: MTF, Monte Carlo
Poster panel: 64

Poster Number:

Extension of the List-Mode MLEM algorithm for poly-energetic imaging with a Compton Camera (#1292)

B. Mehadji1, M. Dupont1, Y. Boursier1, C. Morel1

1 Aix-Marseille Univ, CNRS/IN2P3, CPPM, MARSEILLE, France

This work was supported by the project TEMPORAL funded by the ANDRA/PIA under the grant No. RTSCNADAA160019.


We present an extension of the List-Mode Maximum Likelihood Expectation Maximization (LM-MLEM) algorithm for Poly-Energetic imaging with a Compton camera: PE-LM-MLEM. A reasonable estimation of the geometrical acceptance of the Compton camera is achieved by using Gate Monte Carlo simulations, which allows us to reconstruct successfully radioactive source activities and energies of simulated data. We present reconstructed poly-energetic images and  highlight some limitations of the LM-MLEM algorithm in terms of image correction based on the sensitivity matrix of the Compton camera.

Keywords: Compton camera, poly-energetic imaging, LM-MLEM, PE-LM-MLEM, GATE
Poster panel: 67

Poster Number:

Comparison of noise equivalent count rates (NECRs) for small diameter brain PET system and whole-body PET system (#1552)

K. Nakanishi1, Y. Hirano2, S. Yamamoto2

1 Nagoya University Graduate School of Medicine, Nagoya, Japan
2 Nagoya University, Graduate School of Medicine, Nagoya, Japan


Because the sensitivity of a PET system increases in proportion to the inverse of the diameter of the detector ring, small-diameter brain PET systems are developed or being planned. However, count loss due to the dead-time of the detector blocks increases as the ring diameter decreases. Since the count loss decreases the noise equivalent count rate (NECR) of the system, the NECR of a small-diameter brain PET system may be lower than that of a whole-body PET system in a dose range used for clinical examination. In this paper, we compared the NECRs of brain PET systems with that of a whole-body PET system using a Monte Carlo simulation. We simulated the PET system with a 26-cm-diameter detector ring, a 52-cm-diameter detector ring and a 78-cm-diameter detector ring and evaluated the NECRs for two types of 20-cm-diameter cylindrical phantoms with different axial lengths of 15 cm and 70 cm. In the clinical dose range, the whole-body PET system could achieve an NECR several times higher than that of the small-diameter brain PET system. We conclude that the small-diameter brain PET system has little advantage over a whole-body PET system for clinical studies except for its lower cost.

Keywords: NECRs, simulation
Poster panel: 70

Poster Number:

The effects of multiple interactions in scintillation crystals on the performance of PET (#1789)

C. Zhang1, Z. Kuang1, Q. Yang1, X. Wang1, X. Fu1, J. Gao1, N. Ren1, Z. Sang1, X. Zhang1, S. Wu1, Z. Hu1, Y. Zheng1, B. Zhao1, J. Du1, Y. Yang1

1 Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China


Multiple interactions in scintillator crystal affect the spatial resolution of PET scanners since either one or both gamma rays undergo multiple interactions for most of the events. In this work, Monte Carlo Simulations were performed to study the effects of multiple interactions in scintillator crystal on the sensitivity and spatial resolution of PET scanners. LaBr, LYSO and PWO that representing scintillator crystal of low, medium and high density were used. The positron range and photon non-collinearity effects are not considered in all simulations. For two scintillator detectors of  50×50×20 mm3 and a lower energy threshold of 350 keV, the probabilities that at least one gamma ray undergoes multiple interactions are 94%, 94% and 76% for LaBr, LYSO and PWO, respectively. The multiple interaction events still provide useful spatial information. The FWHM and FWTM of the curves of the mis-positioning distribution caused by multiple interaction are 0.5 and 3.7 mm if the most popular PET scintillator LYSO is used.  The FWHM is much smaller than the spatial resolution of most PET scanners. Multiple interaction events have relative bigger effects on FWTM of the spatial rsolution. The coincidence counts between one column of the crystals and the sum of the coincidence counts between two opposite crystals in the two detectors are used to measure the intrinsic spatial resolution of a pair of LYSO detectors. The latter method removes lots of multiple interaction events and is expected to provide better resolution. The FWHM intrinsic spatial resolutions obtained by the two methods are 0.40 and 0.33 mm for a crystal size of 0.5 mm. The intrinsic spatial resolutions obtained by the two methods are 0.8 and 0.68 mm if the crystal size is 1.0 mm.

Poster panel: 73

Poster Number:

Calculation and impact of the sensitivity model for a two-plane Compton camera (#1878)

E. Muñoz1, L. Barrientos1, J. Barrio1, M. Borja-Lloret1, A. Etxebeste1, L. Gabarda1, C. Lacasta1, G. Llosá1, A. Ros García1, J. Roser1, J. Oliver1

1 Instituto de Física Corpuscular, Paterna, Valencia, Spain


Given the strong variations in the sensitivity of Compton cameras for the detection of events originating from different points in the field of view (FoV), sensitivity correction is often necessary in Compton image reconstruction. Several approaches for the calculation of the sensitivity matrix have been proposed in the literature. While most of these models are easily implemented and can be useful in many cases, they usually assume high angular coverage over the scattered photon, which is not the case for our prototype. In this work, we propose an analytical model that allows us to calculate a detailed sensitivity matrix, and compare it to other sensitivity models in the literature. The sensitivity model has been validated taking Monte Carlo simulations as a reference. In order to study the impact of the sensitivity, images reconstructed with our sensitivity model and with other models have been compared. Images have been reconstructed from several simulated sources, including extended distributions of activity, and also from experimental data measured with Na-22 sources.

Results show that our model allows to effectively recover the intensity of point-like sources at different positions in the FoV and to reconstruct regions of homogeneous activity with minimal variance. Moreover, it can be employed for all Compton camera configurations, including those with low angular coverage over the scatterer.

Keywords: Compton Imaging, Hadron therapy
Poster panel: 76

Poster Number:

Design study of wearable high performance brain PETs (#2044)

W. Tao1, F. Weng1, G. Chen1, Y. Zan1, Q. Peng3, J. Xu2, Q. Huang1

1 Shanghai Jiao Tong University, Shanghai, China
2 Huazhong University of Science and Technology, Wuhan, China
3 Lawrence Berkeley National Laboratory, Berkeley, California, United States of America


The purpose of this study is to investigate the feasibility of building a wearable brain PET with a full-brain coverage (aperture > 24 cm, axial length > 12 cm) and an effective sensitivity approaching the commercial dedicated brain scanner (HHRT) and the whole-body scanner (mCT) using a limited volume of LYSO scintillator crystals (weight < 2.5 kg).  Six typical designs with different discrete LYSO scintillator crystal sizes, which are the combinations of 2 different crystal pitches (3 mm and 1.5 mm) and 3 different crystal lengths (20 mm, 10 mm and 5mm), were simulated. Sensitivity and image noise property of the six designs were evaluated and compared with those of the HRRT scanner and mCT scanner. For SNR assessment, the list mode data were reconstructed with 5 different Time-of-Flight (TOF) settings (non-TOF, 500 ps, 200 ps, 100 ps and 50 ps Coincidence Time Resolution, CTR) using the maximum likelihood expectation maximization (MLEM) algorithm. The results show that a fully wearable brain PET with the 50 ps CTR is able to achieve excellent performances comparable with the HRRT system (without TOF) and the mCT system (with the TOF resolution being 527 ps). The PET scanner design with four detector rings, each consisting of 19 detector blocks, each block containing an array of 20×20 crystals (1.5 mm×1.5 mm×5 mm) is a good choice. The system has a diameter of 237 mm and an axial FOV of 130 mm. The total crystal weight is no more than 2.5 kg.

Keywords: Design study, weraable brain PET, GATE
Poster panel: 79

Poster Number:

Grid Design for Gamma-ray Computed Tomography for Explosive Detection: A Simulation Study   (#2185)

E. Pratiwi1, S. Bae2, H. Lee2, B. Lee3, K. Lee1, 4, J. - Y. Yeom1, 4

1 Korea University, Bio-Convergence Engineering, Seoul, Republic of Korea
2 Arale Laboratory, Seoul, Republic of Korea
3 Samsung Medical Center, Seoul, Republic of Korea
4 Korea University, School of Biomedical Engineering, Seoul, Republic of Korea



Abstract—In this simulation study, two type of grids were used to enhance the spatial resolution of gamma CT arch detector for explosive detection system. Both grids were made of tungsten, with same length (30 mm), but different hole shapes: Fan and Parallel. In both cases, we could see the improvement clearly, especially in smaller rods. The peak-to-valley (PV) ratio improved from ~0.25 to more than 0.4. And for the grids themselves, Fan grid performed better than Parallel grid in most cases (PV increased up to 0.1).

Keywords: gamma CT, collimator, simulation, arc detectors, grid
Poster panel: 82

Poster Number:

Preliminary Monte Carlo simulations of a SPECT system based on CdZnTe detectors for real time BNCT dose monitoring (#2260)

N. Protti1, S. Fatemi1, C. Gong2, S. Bortolussi1, 3, C. Magni4, I. Postuma1, X. Tang2, S. Altieri1, 3

1 National Institute of Nuclear Physics INFN, Pavia Unit, Pavia, Italy
2 Nanjing University of Aeronautics and Astronautics, Nanjing University, Nanjing, China
3 University of Pavia, Department of Physics, Pavia, Italy
4 National Institute of Nuclear Physics, Milano Unit, Milano, Italy


Boron Neutron Capture Therapy (BNCT) is a binary radiation therapy which is able to selectively destroy malignant cells while sparing the normal tissue. A 10B containing drug able to target neoplastic cells is administered to the patient which is then irradiated with thermal neutrons that induce the 10B(n,α)7Li capture reaction. Therefore BNCT effectiveness depends on the ability to induce a high concentration of 10B inside the neoplastic tissue. The correct evaluation of such quantity and of the dose deposited in the tumour is a key element to further BNCT efficacy. The current methods clinically used to estimate the boron concentration are indirect. To obtain a direct and real time quantification of the dose delivered to the tumour a BNCT-SPECT system has been proposed. SPECT imaging in BNCT is based on the 478 keV photon emitted in the 94% of the cases due to the 10B thermal neutron capture.

To develop a BNCT-SPECT imaging system a CdZnTe (CZT) detector was chosen to have, even with small sensitive volumes, high efficiency, good energy resolution and possibility of working at room temperature.

The present work focused on the simulations of a 20x20x20 mm3 CZT detector as the base element of the named SPECT. The system was first simulated in an ideal setup using a virtual collimator of 1 mm aperture obtaining 7 mm spatial resolution. Then a more realistic cases was simulated using different collimators and multiple sources. The latter shows a different spatial resolution that could be improved by using a bigger sensitive volume and using smaller angular steps for the data acquisition, but it is still possible to reconstruct the image of the 478 keV source and to discriminate multiple sources within a phantom.

This work shows the capabilities of a small 20x20x20 mm3 CZT detector as a SPECT imager in BNCT and also that with some improvements it should be possible to use a CZT detector as a base element for a BNCT-SPECT imaging system.

Keywords: CZT detector, BNCT, Monte Carlo simulation, SPECT
Poster panel: 85

Poster Number:

Orientation Dependent Visualization of Fibers in Digital Breast Tomosynthesis: Advantages of a Circular Source Trajectory (#2377)

S. Rose1, I. Reiser1, E. Y. Sidky1, X. Pan1

1 University of Chicago, Radiology, Chicago, Illinois, United States of America


Visualization of fiber-like structures is important in the context of breast imaging. Cooper's ligaments, vessels, ducts, and spiculations from potentially malignant lesions are all examples of fiber-like features that can be present in the breast. In digital breast tomosynthesis, fiber-like signals appear more conspicuous when aligned perpendicular to the direction of X-ray source travel than when they lie parallel. This in-plane anisotropy results from the use of an arc-shaped source trajectory lying in a plane perpendicular to the detector. In this work, we investigate the use of a circular X-ray source trajectory to mitigate the anisotropic visualization of fiber-like signals. A preliminary simulation study is performed to characterize the orientation-dependence of fiber-like signal conspicuity using both arc-shaped and circular source trajectories. The impact of regularization strength employed in reconstruction is also investigated in both geometries. Reconstructions from the arc-shaped trajectory are seen to exhibit orientation-dependent conspicuity and/or depth blur depending on the regularization strength. Using the circular trajectory, these orientation-dependent phenomena are not observed at any regularization strength.

Keywords: breast imaging, digital breast tomosynthesis, iterative image reconstruction
Poster panel: 88

Poster Number:

Modular Detector Designs for Multi-Head Variable Pinhole SPECT (#2412)

S. Bae1, H. Cha2, A. Santos3, K. M. Kim4, K. Lee2, H. Lee1, 5

1 ARALE Laboratory, Inc., Seoul, Republic of Korea
2 Korea University, Department of Bio-convergence Engineering, Seoul, Republic of Korea
3 Korea University, School of Biomedical Engineering, Seoul, Republic of Korea
4 KIRAMS, Radiation Therapy System Development Team, Seoul, Republic of Korea
5 Korea University, Research Institute of Global Health Tech., Seoul, Republic of Korea


Recently, a novel SPECT system which can change the sensitivity and spatial resolution parameters for each rotation angle in real-time, has been announced. This system utilizes a variable pinhole(VP) collimator that can be optimized for each patient-oriented target organ. The superior performance of the preclinical VP SPECT has been explored through simulation studies.
In order to maximize and enlarge performance, the maximum object size of VP SPECT, a newer version of the system, which employs 24 compact SPECT heads with a stationary gantry is under development. This proposed system is designed for the study of the human brain.
In this experiment, we explored the artifacts of scintillator gap between tiled detector module for the multi-head VP SPECT. To reduce the detector gap and depth-of-interaction(DOI) artifact near edge of the outside detectors, we proposed an arc-shaped tiled detector design. The initial performances of this proposed detector shape were studied by Monte Carlo simulation.
As a result, for each of flat and arc-shaped detector cases, full width at half maximum(FWHM) of the PSF at central isotope, was measured similarly to 6.38 mm and 6.15 mm, respectively, while each of FHWM was measured as 6.59 mm and 5.70 mm at the edge positioned isotope.
The performance of two different designs for tiled detectors has been studied. Those respective results aid in developing a cost effective and large sized gamma detector with compact modular detectors.
We are currently working on developing a gap compensation algorithm for this specific system. Therefore, the performance of proposed SPECT system will be analyzed quantitatively.

Keywords: SPECT, Variable Pinhole, Modular Detector, Image Reconstruction
Poster panel: 91

Poster Number:

Preliminary investigation of design parameters of an innovative multi-pinhole system dedicated to brain SPECT imaging (#2454)

B. Auer1, J. De Beenhouwer2, T. J. Fromme3, K. S. Kalluri1, J. C. Goding1, G. I. Zubal4, L. R. Furenlid5, M. A. King1

1 University of Massachusetts Medical School, Dept. of Radiology, Worcester, Massachusetts, United States of America
2 University of Antwerp, imec-VisionLab, Antwerp, Belgium
3 Worcester Polytechnic Institute, Robotics Engineering, Worcester, Massachusetts, United States of America
4 Z-Concepts, LLC, East Haven, Connecticut, United States of America
5 University of Arizona, Dept. of Radiology & The College of Optical Sciences, Tucson, Arizona, United States of America


Collimator penetration, down-scatter related to 123I high energy photons, and scatter within detectors can significantly degrade the imaging performance of any system. Precise selection of pinhole and collimator parameters using simulation studies has the potential to considerably reduce these effects. This type of investigation, thus represents an essential step in system development. An innovative multi-pinhole system, AdaptiSPECT-C, dedicated to clinical brain SPECT imaging, is currently under development at the universities of Massachusetts and Arizona.
The aim of this work was to determine the system parameters which considerably improve AdaptiSPECT-C imaging performance for the criteria of sensitivity and relative amounts of scatter and penetration.
The need for internal shielding was clearly highlighted. A shielding thickness of 5 mm has the potential to significantly decrease the amount of inter-module interactions (up to 90%). Shielding materials composed of Cerrobend or lead alloy represent effective alternatives to tungsten alloy at less cost for inter-module shielding. A 20 mm thick, tungsten alloy collimator leads to the best trade-off between performance and price in case of 123I imaging. Tungsten alloy provided performance relatively close to that of gold in terms of stopping power as compared to lead alloy. A pinhole center distance of 0.5 cm to the aperture entry port led to the best compromise for locating the aperture within the aperture plate in terms of sensitivity and relative amounts of scatter and penetration. A keel edge design of 1 mm height appeared to be a good alternative to a knife edge pinhole, as the relative amount of penetration could be reduced by a factor of two but at the expense of sensitivity (-17%).

Keywords: 123I SPECT imaging, next-generation clinical system, Optimization study, effects degrading imaging performance
Poster panel: 94

Poster Number:

Investigation of the impact of novel scintillation crystals on the performance of PET scanners (#2588)

A. Ghabrial1, A. Franklin2, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Genève, Switzerland
2 University of Technology Sydney, Sydney, Australia


The purpose of this study is to validate a Monte Carlo simulation model for the clinical Siemens Biograph mCT PET scanner using the GATE simulation toolkit, and to evaluate the performance of six different scintillation materials in this model using the National Electrical Manufactures Association (NEMA) NU 2-2007 protocol. A model of the Biograph mCT PET detection system and its geometry was developed. NEMA NU 2-2007 phantoms were also modelled. The accuracy of the developed scanner model was validated through a comparison of the simulation results from GATE, SimSET and PeneloPET toolkits, and experimental data obtained using the NEMA NU 2-2007 protocols. The evaluated performance metrics included count rate performance, spatial resolution, sensitivity, and scatter fraction (SF). Thereafter, the mCT PET scanner was simulated with six different candidate high-performance scintillation materials, including LSO, LaBr3, CeBr3, LuAP, GLuGAG and LFS-3, and its performance evaluated according to the NEMA NU 2-2007 specifications. The Monte Carlo simulation model demonstrates good agreement with the experimental data and results from other simulation packages. For instance, the scatter fraction calculated using GATE simulation is 34.35% while the experimentally measured value is 33.2%, 38.48% for the SimSET, and 34.8% for the PeneloPET toolkit. The best-performing scintillation materials were found to be LuAP, LSO and LFS-3, while GLuGAG offers acceptable performance if cost is the dominant concern. The main performance characteristics of the Biograph mCT PET scanner can be simulated accurately using GATE with a good agreement with other Monte Carlo simulation packages and experimental measurements. Newly developed scintillators show promise and offer alternative options for the design of novel generation PET scanners.

Poster panel: 97

Poster Number:

Dedicated Cardiac SPECT for diagnostic and theranostics applications (#2648)

S. Sajedi1, L. Bläckberg1, 2, G. El Fakhri1, H. Sabet1

1 Massachusetts General Hospital & Harvard Medical School, Radiology, Charlestown, Massachusetts, United States of America
2 Uppsala University, Physics and Astronomy, Uppsala, Sweden


 The primary limitation in all cardiac SPECT systems is their low photon detection efficiency mainly due to low radiotracer concentration in the myocardium and use of collimators. This partly explains the relatively poor image resolution of current cardiac SPECT systems. Moreover, with the development of new fast-flowing radiotracers, the need for systems with fast acquisition time is more pronounced. Therefore, all new SPECT systems dedicated to cardiac application seek to provide faster acquisition, and their geometries are designed specifically to focus on the heart region. We are developing a stationary cardiac SPECT system that can provide high-resolution images reaching far superior to any existing solution and ~8-10x sensitivity improvement over the conventional dual head gamma cameras. Unlike the two state-of-the-art commercial systems, GE’s NM-530 and Spectrum-Dynamic’s D-SPECT, that are based on CZT detector modules, our design is based on scintillators with high intrinsic detector resolution (IDR) pixelated by using laser-induced optical barriers technique. The current design of our cardiac SPECT is comprised of 80, 50x50 mm2 CsI:Tl detector modules with 2x2x10 mm3 pixels and their associated 2.5x2.5 mm2 loft-hole collimator. The detector modules are arranged in 4 rows and 20 detector modules per row covering ~225c detector arc which provides large number of simultaneous view samples. The collimator to CFOV varies between 15.8 to 24.3 cm depending on the detector module location with respect to the FOV with an average 20.5 cm distance. Our initial simulation results show that the proposed system provides 0.079% system geometric sensitivity at 10.5 mm FWHM system spatial resolution (SR) without resolution recovery at 20.5 cm average collimator-CFOV distance. This translates to ~8x improvement in sensitivity compared with dual head gamma cameras encouraging a lower patient dose, larger patient throughput, or the possibility of dynamic cardiac studies.

Keywords: SPECT, Cardiac, CsI:Tl, high-resolution, laser
Poster panel: 100

Poster Number:

Performance evaluation of a Compton SPECT imager for position and distribution of 225Ac in targeted alpha therapy: Monte Carlo simulation phantom study (#2757)

T. Lee1, M. Kim1, B. S. Kim1, I. Lim1, K. Song1, W. Lee2, J. Kim1

1 Korea Institute of Radiological and Medical Sciences, RI Applied Research Team, SEOUL, Republic of Korea
2 Korea University, School of health and Environmental Science, SEOUL, Republic of Korea


The purpose of this study is to compare and evaluate the performance of a Compton single-photon-emission-computed-tomography (SPECT) to monitor γ-rays emitted during ‘Targeted Alpha Therapy’ (TAT) using 225Ac pharmaceuticals. 225Ac decays to stable 209Bi via two beta and four alpha decays. In the decay chain, both the 218- and 440-keV γ-rays emitted by the 225Ac progeny 221Fr and 213Bi could be imaged. The Compton SPECT has an area of 30 cm 30 cm with four gantry heads. Each head is composed of a 3 cm tungsten collimator and a 100 100 array of virtual Frisch-grid CdZnTe (CZT) crystals with 3 3 mm2 area and 6-mm-thickness. The Compton SPECT can use not only the photoelectric events but also the Compton scattering events for image reconstruction. Therefore, the detection efficiency and quality of the reconstructed image can be improved significantly. Under the correct sequential order and proper energy window for both photoelectric and Compton scattering events, all effective information from the γ-rays of 225Ac whose energies 218- and 440 keV was utilized in the reconstructed image. Finally, the contrast-noise ratio, spatial resolution and uniformity of Compton SPECT were evaluated and compared with that of the conventional SPECT using photoelectric events only. The quantitative results prove that the performance of the Compton SPECT is better than that of the conventional SPECT.

Keywords: Compton SPECT, Ac-225, CZT detector
Poster panel: 103

Poster Number:

Dual-ended readout small animal PET detectors with 0.5 mm pixelated LYSO crystal arrays and SiPMs (#1123)

Z. Kuang1, X. Fu1, X. Wang1, N. Ren1, Q. Yang1, B. Zhao1, C. Zhang1, S. Wu1, Z. Sang1, Z. Hu1, J. Du1, Y. Yang1

1 Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China


Three 17×17 pixelated LYSO arrays with crystal size of ~0.5 mm and an outer dimension of 10×10×20 mm3 are fabricated. The first LYSO array uses 80 μm thick BaSO4 reflector. The second and the third LYSO arrays use 50 μm thick Toray reflectors. The outermost of the third LYSO array was also wrapped with k9 glass of the same size as the crystals in the array to improve the light collection and flood histogram for the edge crystals. The LYSO arrays were read out with two Hamamatsu SiPM arrays from both ends. The SiPM array is 4×4, with a pixel size of 3×3 mm2 and 0.2 mm gap in between. The performance of the detector was measured. All three detectors provide good flood histograms and all crystals can be clearly resolved from the measured flood histograms. The average DOI resolutions of individual crystals are 1.87, 1.98 and 1.93 mm for the three detectors for events with E>350 keV. The BaSO4 array has higher light output and provides a better energy resolution of 18.4%. The photopeak amplitude of the Toray arrays changes with depth. The average energy resolution of the two detectors are 24.2% and 23.6% respectively. All three detector modules can be used to build a high sensitivity small animal PET scanner with spatial resolution approaching the physical limit.

Keywords: depth of interaction; small animal PET; dual-ended readout; SiPM; physical limit
Poster panel: 106

Poster Number:

Optimization of a Light Guide for High Resolution PET Detectors using GATE Optical Simulation  (#1390)

H. G. Kang1, S. Takyu1, F. Nishikido1, A. Mohammadi1, N. Inadama1, E. Yoshida1, T. Yamaya1

1 National Institute of Radiological Sciences (NIRS) in National Institutes for Quantum and Radiological Science and Technology (QST), Department of Radiation Measurement and Dose Assessment, Chiba, Japan


A small animal positron emission tomography (PET) scanner employs a scintillation crystal with a small cross section to improve the spatial resolution. However, the separation of crystals becomes more difficult as the cross section of the crystal is reduced. In particular, the identification of edge crystals is more difficult than the central crystals. The aim of this study is to optimize a light guide design of a small animal PET detector to improve the resolvability of the edge crystals using GATEv6.2 optical simulation. The small animal PET module consists of an 11 × 11 array of pixelated LYSO crystals (0.92×0.92×10.0 mm3), a 1 mm thick light guide, and a 4 × 4 array SiPM (Hamamatsu Photonics, Japan; S13361-3050NE-04; each channel has an effective area of 3 × 3 mm2). The optical characteristics of LYSO and the SiPM were taken into account for the GATEv6.2 optical simulation. In order to improve the crystal resolvability in the edge and corner regions, four air slits were inserted into the light guide. The flood map was obtained with various slit depths of 0.2, 0.5, and 0.8 mm, respectively. The separation of the corner crystals could be improved by using a combination of a light guide thickness of 1 mm, and a slit depth of 0.5 mm without compromising energy resolution. In future, the optical simulation results will be validated by an experimental measurement.

Keywords: GATE optical simulation, light guide
Poster panel: 109

Poster Number:

Depth-of-Interaction Encoding Detectors for the SAFIR System (#1723)

M. Ito1, C. Tsoumpas2

1 ETH Zurich, Institute of Particle Physics, Geneva 23, Switzerland
2 University of Leeds, Leeds Institute of Cardiovascular and Metabolic Medicine, Leeds , United Kingdom

On behalf of SAFIR-Collaboration


The SAFIR (Small Animal Fast Insert for mRi) system is a high-rate PET insert for a 7T MRI to study kinetics of short-lived tracers. In order to handle high rate data, the detector should be designed to minimize pile-up effect. Excellent timing performance is also required for using a narrow coincidence timing window to reduce random contributions.

In this study, three different depth-of-interaction (DOI) encoding detectors were evaluated for the SAFIR system to improve spatial resolution: i) a dual-layer offset, ii) a high resolution (HR) dual-layer offset, and iii) a single-layer DOI detector with triangular reflectors. The three DOI-encoding crystal arrays are mounted on a 4 × 4 SiPM array with 2 mm × 2 mm pixels.

The dual-layer offset detectors provide discrete DOI information, and all crystals in the two layers were well separated in flood image. The peak-to-valley ratio was 9:1 for dual-layer offset and 4:1 for HR dual-layer offset. The single-layer DOI detector with triangular reflectors provides continuous DOI information by measuring difference of light diffusion depending on the DOI. The average DOI resolution was calculated to be ~3.1 mm FWHM. We report the energy and timing performances of the DOI-encoding detectors as compared with a non-DOI detector composed of the same dimensions of crystals and SiPMs with one-to-one coupling (energy resolution of 12.1% and coincidence resolving time (CRT) resolution of 319 ps FWHM). The dual-layer offset detectors provide better energy resolution of ~11.4% for both layers, but lead to ~80 ps and ~60 ps degradations in the timing performance for the top and bottom layers, respectively. The single-layer DOI detector exhibits not only better energy performance of 10.2%, but also ~25 ps better timing resolution than a non-DOI one-to-one coupling detector.

Keywords: SAFIR, pre-clinical PET/MRI, Depth-of-interaction (DOI), detector performance
Poster panel: 112

Poster Number:

Characterization of a scalable detector stack for preclinical PET/MR (#1672)

A. R. Selfridge1, A. Kolb1, 2, R. D. Badawi2, 1, S. R. Cherry1, 2, M. S. Judenhofer1, 2

1 University of California, Davis, Biomedical Engineering, Davis, California, United States of America
2 University of California, Davis, Radiology, Sacramento, California, United States of America


We have developed a prototype detector stack for a high-sensitivity preclinical PET/MR insert. The detector uses a 2 cm thick array of 19x19 1.0 mm pitch LYSO crystals to maximize sensitivity, with dual-ended DOI encoding readout to improve homogeneity of spatial resolution across the system field-of-view. The detector stack is built with discrete components and read out with a Xilinx Artix-7 FPGA for a relatively low cost without a dedicated ASIC. The readout electronics are designed around a high-density sigma-delta ADC, modified to provide accurate signal digitization for highly multiplexed photodetector. A single eight-channel readout board is sufficient to digitize the block detector. The complete acquisition interfaces with a workstation over a USB connection.

The combination of our prototype detector module and readout stack resolves all 19x19 crystal elements with 15 +/- 11% energy resolution, 3.7 +/- 1.3 mm depth of interactions resolution, and 2.0 +/- 0.2 ns timing resolution. We are currently designing the final detector stack to meet the space requirements of the proposed PET/MR insert, providing dual-ended readout of 64 blocks within the 20 cm bore of the 7T Bruker BioSpec preclinical MRI, leaving a 9 cm diameter for imaging subjects and gradient coils. This compact stack design will place all processing electronics within the bore of the MRI, only requiring cabling to deliver power and transfer data to an acquisition computer.

Keywords: sigma delta, positron emission tomography, PET/MRI, depth of interaction
Poster panel: 115

Poster Number:

Feasibility study and pilot results of a PET system based on a single LYSO tube (#1753)

G. Cañizares1, S. Berr2, A. González-Montoro1, A. Orero3, C. Correcher3, S. Majewski2, A. Rezaei4, J. Nuyts4, J. M. Benlloch1, A. J. González1

1 Universitat Politècnica de València, Instituto de Instrumentación para Imagen Molecular (I3M), Valencia, Spain
2 University of Virginia, Charlottesville, Virginia, United States of America
3 Bruker NMI, Valencia, Spain
4 KU Leuven, Nuclear Medicine and Medical Imaging Research Center, Leuven, Belgium


Most commercially available PET scanners are based on pixelated technology. As an alternative, monolithic scintillators have been employed also providing photon DOI. 3D absorption localization for both pixelated and continuous crystals is inaccurate at the edges, leading to a degradation of the spatial resolution. Spatial resolution will improve if the edges of the scintillator are removed.

We acquired data using continuous scintillators and compared the resolution in the reconstructed image when omitting coincidences near crystal edges, simulating the proposed continuous-tube detector behavior. The data set was acquired using a mini Derenzo (0.75 mm) imaged with a prototype Bruker small animal PET insert. This scanner employs continuous crystals (10 mm thickness).Projections of reconstructed data through the hot rods were obtained. A fitted Gaussian width of 1.45 mm was determined for the Original data set and 1.19 mm for the Filtered data set. This represents a roughly 23% increase in resolution. Based on these results, the resolution of the proposed system is estimated to be 0.6 mm.

We propose a design based on a single monolithic 80 mm outer diameter LYSO tube, with 60 mm inner diameter, and about 80 mm axial distance. We introduce the outer faces to be facetted for designing and SiPM coupling simplicity. We have simulated both nuclear and optically this design with good results. We have also reconstructed data not showing any deterioration when compared to standard multiple blocks-based data. Light distributions exhibit the system capability to resolve well the 3D impacts coordinates, X, Y and DOI.

A PET scanner built around an ‘edgeless’ scintillator tube design is expected to have a spatial resolution of 0.6 mm, which approaches the spatial resolution limits of PET. We will show a design based on a single tube with an inner circular face and 10 facetted outer faces. Simulations and reconstruction data is currently undergoing.

Keywords: Small animal PET, monolithic crystals, hybrid PET-MR, scintillation tube
Poster panel: 118

Poster Number:

Towards a Next-Generation PET/MRI Insert with Improved Timing and Count Rate Performance (#1965)

D. J. Van Elburg1, A. L. Goertzen1, 2

1 University of Manitoba, Department of Physics and Astronomy, Winnipeg, Manitoba, Canada
2 University of Manitoba, Department of Radiology, Winnipeg, Manitoba, Canada


We have previously developed a first generation PET insert prototype for PET/MR imaging of small animals, with a detector design utilizing SensL B-series SiPMs readout using a resistor-based charge division multiplexing network. The count rate performance of this system was limited by a 2.5 ns coincidence time resolution and a ~1.3 μs paralyzable deadtime. In this work, we explore detector designs and optimized signal capture for a 2nd generation insert with improved performance. A 2nd generation detector board with 4×8 SiPM array support was developed, which retains the resistor-based multiplexing network but is optimized for improved count rate performance. Two variants of this board were built and tested: i) using SensL C-series Array-30035-16P-PCB (SiPM-C), including support for the ‘fast’ outputs available on these SiPMs; and ii) using Hamamatsu S11361-3050AE-04 (MPPC) arrays. Timing data were obtained in coincidence with a single SensL 3 mm J-series SiPM pixel using standard NIM electronics. Count rate data were obtained using the OpenPET data acquisition system. For SiPM-B, SiPM-C standard output, SiPM-C fast output, and MPPC, timing resolutions were 1847 ± 141 ps, 806 ± 86 ps, 425 ± 33 ps, and 435 ± 26 ps respectively, while pulse duration (time above 10% amplitude) was 1600 ns, 560 ns, 560 ns, and 290 ns respectively. Count rate data were obtained using 68Ge rod sources at distances varied from 0-90 mm. Pileup effects were observed in flood images of SiPM-B and SiPM-C at 281 kcps and 565 kcps count rates respectively, while MPPC flood images showed negligible pileup, even at 569 kcps. The results suggest that the 2nd generation detector design will provide significant improvements in count rate performance through reduced deadtime and increased stability vs. event rate compared to the first generation design.

Keywords: Positron emission tomography, Time-of-flight (ToF) PET, PET insert, SiPM, MPPC
Poster panel: 121

Poster Number:

A Novel Radionuclide Endoscopic Imaging System for Hepatocellular Carcinoma Guided Resection on Murine Models (#2045)

Z. Zhang1, 2, M. Cai2, Z. Hu2, J. Tian2

1 Xidian University, School of Life Science and Technology, Xi’an, China
2 Chinese Academy of Sciences, Institute of Automation, Key Laboratory of Molecular Imaging, Beijing, China


Cerenkov luminescence imaging (CLI) has been proved to be an effective imaging modality for biomedical application. A series of studies were conducted to strengthen the signal intensity with quantum dots (QDs) or inorganic nanoparticles, which strongly obstacle the clinical translation. In this study, we developed a novel endoscopic imaging system for Cerenkov radiation energy transfer imaging (CRET) with a clinical laparoscope, which integrated air-liquid dual cooling solution to improve the imaging sensitivity. FDA approved sodium fluorescein was applied in our study to enhance the CL signals for raising the potential of clinical translation.

Hepatocellular carcinoma (HCC) murine models were used for in vivo experiments of CRET tumor detection and guided-resection. Then, in vivo intraoperative endoscopic CRET imaging was performed to guide the hepatocellular carcinoma (HCC) resection. With guidance of the endoscopic CRET images, we succeeded to resect the tumor tissues on murine models. The results indicates that: (1) successful tumor resection was able to achieve by the guidance of endoscopic CRET; (2) endoscopic CRET can acquire the optical signals from deep orthotropic HCC, but CLI cannot; (3) by using of sodium fluorescein, the optical signals were clearly enhanced and believed feasible to detect the early-stage small tumors in improved sensitivity and reveal the tumor margins with high SBR. Postoperative pathological results verified feasibility and resection performances of the endoscopic CRET system. We believed the endoscopic CRET system is promising to increase the sensitivity for early stage tumor detection with great potential in clinical translation.

Keywords: Endoscopy System, Hepatocellular carcinoma, Cerenkov radiation energy transfer imaging (CRET)
Poster panel: 124

Poster Number:

Preliminary result of a SiPM based DOI-PET detector for a small animal PET using depth-dependent reflector pattern (#2098)

H. Song1, C. Park1, I. S. Kang1, M. K. Baek1, G. Lee1, K. B. Kim1, Y. H. Chung1

1 Yonsei University, Department of Radiation Convergence Engineering, Wonju-si, Republic of Korea


In this study, we have developed a SiPM based DOI-PET detector modules using a depth-dependent reflector patterns within a single layer scintillator. The scintillators were painted by reflector material, a titanium dioxide paint (BC-620, Saint Gobain) to distinguish DOI according to the crystal depth. The heights of the front layer and back layer are 4 mm and 16 mm, respectively, to get the equivalent detection efficiency. The crystal block consists of 12×12 LYSO crystals with 2.1×2.1×20 mm3 size. The crystal block is optically coupled to the 8×8 SensL SiPM array with a pixel size of 3×3 mm2 and a pixel pitch of 3.36 mm. A 2-D histogram was obtained using a Na-22 source and all two layers of 12×12 LYSO pixels were clearly identified. The average peak to valley ratio was 3.52 ± 0.92 and 3.80 ± 0.87 at the front and back layer, respectively. The average energy resolution was 18.25 ± 3.13% and 18.47 ± 3.36 for the front and back layer, respectively. An intrinsic spatial resolution and coincidence timing resolution were measured, as well. Based on the results, a full ring PET system will be developed in further study.

Keywords: PET, SiPM, DOI
Poster panel: 127

Poster Number:

Design and performance study of a quasi-spherical PET scanner and hexagonal SiPM (#2345)

D. Pérez-Benito1, R. Chil1, J. L. Herraiz2, 3, G. D. Konstantinou1, J. M. Udias2, M. Desco1, 4, J. J. Vaquero1, 4

1 Universidad Carlos III de Madrid, Bioengineering and Aerospace Engineering, Leganés, Spain
2 Universidad Complutense de Madrid, Grupo de Física Nuclear and UPARCOS, Madrid, Spain
3 Clinical Hospital San Carlos, Institute of Medical Research, Madrid, Spain
4 Instituto de Investigación Sanitaria Gregorio Marañón, Madrid, Spain


Current challenges on new Positron Emission Tomography (PET) scanner design are high sensitivity, spatial and temporal resolution, all aiming to accurately quantify dynamic biological processes. The optimal scanner geometry that maximizes the sensitivity and spatial resolution is a sphere. However, state of the art of scintillators crystals and position-sensitive photodetectors for γ-rays are not easy to arrange in this configuration. In this work we present a proof-of-concept of a preclinical imaging scanner shaped as an icosahedron based on a new hexagonal Multi Pixel Photon Counter (MPPC) Silicon Photomultiplier (SiPM). Each triangular facet (155 mm of side; 234mm diameter of inscribed sphere) has 10 hexagonal 10 mm thick monolithic LYSO laser-engraved scintillator crystals with the Sub-Surface Laser Engraving (SSLE) technique. The SiPM hexagonal matrix consists of 61 individual hexagonal cells of 2.25 mm of edge. Higher spatial resolution is achieved due to the ratio 1:3 of the area of the engraved pixels with respect to the size of the single SiPM cells. The honeycomb arrangement can be argued as the geometry that maximizes the packing fraction. Depth of Interaction information is obtained by engraving different patterns in the scintillator crystal. This setup results in a geometrical efficiency of 73% with respect to a full 4π coverage. Simulated results for Noise Equivalent Count rate (NEC), sensitivity and a first approach to the study of the spatial resolution based on Fully Bayesian Image Estimation (FBI3D) method are presented. This design shows favorable characteristics for fast, dynamic high-resolution total body PET imaging in mice or low-dose brain studies. Anger-like schemes for sectorial readout of the facets are being tested as a first step to increase the system granularity and to reduce dead time.

Keywords: PET, Design optimization, Silicon Photomultipliers, PET scanner, Small animal total body PET
Poster panel: 130

Poster Number:

Investigation of Time-Over-Threshold for a Depth of Interaction PET Detector using the TOFPET2 ASIC (#2561)

D. L. Prout1, A. H. Chatziioannou1

1 University of California, Los Angeles, Crump Institute for Molecular Imaging, Los Angeles, California, United States of America


Towards the development of high sensitivity, high resolution PET scanners, we have constructed a dual LYSO/BGO crystal layer phoswitch block detector. This detector block provides both depth-of-interaction information and cross-layer-crystal-scatter rejection (CLCS) through the delayed-charge-integration (DCI) technique. The current data acquisition system for this detector consists of full waveform digitizing 125 MHz/14 bit ADCs, which ultimately require high multiplexing of the signals to cut down on cost and provide scalability. Here, we are investigating the feasibility of using the PETSys TOFPET2 ASIC with the dual layer detector module. This ASIC would allow far less multiplexing at a reduced cost; however, it only outputs charge integration (QDC), time, and time-over-threshold (ToT) for each channel, and not DCI. In this work, we evaluate if the ToT parameter can be used to identify the LYSO/BGO crystal layers and reject CLCS type events. As a first step we have collected pulse waveforms at 125MHz from single LYSO and BGO crystals mounted on 3x3mm2 SensL SiPMs and processed these using QDC and ToT rather than DCI. In addition, we have used these pulses to create a database, which provides the light output component for a Monte Carlo routine that simulates CLSC events, from which we can gauge how effective ToT is in rejecting such events. We also have preliminary data of QDC and ToT using BGO and LYSO crystals coupled to the TOFPET2 ASIC via single SiPM pixels. The results of the simulations show that it is possible to use the ToT and QDC together to both discriminate between the LYSO and BGO crystal layers and also reject the CLCS. The preliminary results of the measurements indicate the TOFPET2 ASIC may be used with such a dual layer detector; however careful tuning of the thresholds and triggering are required, because of the long decay time encountered with BGO pulses.

Keywords: PET, detector, time-over-threshold, depth-of-interaction, lyso/bgo
Poster panel: 133

Poster Number:

drimPET : Determination of DoI in LYSO crystals and application of a multiplexing resistive circuit for MPPC readout  (#2677)

P. M. M. Correia1, N. O. Romanyshyn1, I. F. Castro1, F. C. Rolo1, J. F. C. D. A. Veloso1

1 Universidade de Aveiro, i3n - Departamento de Física, Aveiro, Portugal


Abstract - The determination of the depth-of-interaction (DoI) of 511 keV gamma photons in scintillator crystals is of great interest in small diameter positron emission tomography (PET) systems, in order to achieve high spatial resolution with good uniformity within the entire field-of-view. drimPET is new concept for DoI determination wherein a single layer ring of 128 LYSO crystals is read out on both ends: on the outer end, each crystal is directly coupled to a SiPM, while on the inner end groups of several 32 crystals are indirectly coupled to only 1 or 2 SiPMs, using intermediate wavelength-shifters (WLS). This results in a simpler and less expensive readout when compared to the typical dual-ended readout method. Also, less components on the inner side of the ring allow more proximity to the FoV and thus increased sensitivity. Proof-of-concept studies were performed on a single detector cell with electronic collimation. The amount of light detected on each side of the crystal and the achievable DOI resolution with drimPET were measured, as well as the energy resolution, for different positions of a 22Na source along the crystal depth. Experimental studies using a multiplexing circuit based in resistive chain configuration readout and GATE simulations of drimPET are presented to evaluate the feasibility of applying the method in full detector rings for a small animal PET system, considering different number and dimension of the wavelength-shifting fibers.

This work was partially supported by project POCI-01-0145-FEDER- 016855 and PTDC/BBB-IMG/4909/2014, and project easyPET CENTRO-01- 0247-FEDER-017823, CENTRO2020, COMPETE, FEDER, POCI and FCT (Lisbon) programs.

Keywords: DOI, wavelength shifters, PET, Pre-clinical, MPPC
Poster panel: 136

Poster Number:

Development and Evaluation of a High-resolution MWPC-based PET Detector Using 10 µm Anode Wires with 1 mm Spacing (#2838)

K. Bolwin1, B. Gerke1, K. P. Schäfers1

1 University of Münster, European Institute for Molecular Imaging (EIMI), Münster, Germany


Spatial resolution of small animal PET system plays a major role in studying molecular processes in vivo. Beside other physical effects, the photon non-collinearity effect degrades reconstructed images. Since this effect is strongly depending on the detector-to-object distance we optimized the detector setup to improve the system performance. Therefore, we have built a prototype PET scanner by minimizing the detector distance. It could be demonstrated that this approach improves the spatial resolution. One of the resolution limiting factors when using multi-wire proportional chamber (MWPC) detectors is the well known wire effect that restricts the spatial resolution in the direction perpendicular to that of the wire. The aim of this study was to develop and evaluate a MWPC detector build on an existing detector concept from a HIDAC small animal PET system. For this prototype detector, the existing HIDAC detector was redesigned using new anode wires, which are smaller in diameter(10μm, instead of 20μm before) with a reduced wire-to-wire distance of 1 mm (1.5 mm before). In this study we will present the newly developed prototype detector and provide initial experimental results on count distribution as it compares to the existing HIDAC detector.The manufacturing of a second module is in progress to obtain a complete coincidence detector. This should further improve the image-derived point source spatial resolution to a value of approximately 0.7 mm FWHM in all directions.     

Keywords: small animal pet, multi-wire proportional chamber, high resolution
Poster panel: 139

Poster Number:

Performance of a depth encoding PET detector module using light sharing and single-ended readout with SiPMs (#1126)

Z. Kuang1, Q. Yang1, X. Wang1, X. Fu1, N. Ren1, C. Zhang1, B. Zhao1, Z. Sang1, Z. Hu1, J. Du1, Y. Yang1

1 Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China


Detectors with depth encoding capability and good timing resolution are required to develop high performance whole body and total body PET scanners. In this work, an 8×8 LYSO crystal array was fabricated with an optimized coupling material pattern between adjacent crystals in one direction. The coupling materials in between the two crystals which were optimized in our previous work include 5 mm optical glue, 9 mm triangle ESR reflector and 6 mm rectangle ESR reflector. The LYSO array was singled-ended readout by an 8×8 SiPM array with one-to-one coupling. Row and column summing circuit was used to read out the SiPM signals. The depth of interaction (DOI) of the detector is obtained by measuring the light sharing between the two adjacent crystals. Unpolished LYSO crystals of 3×3×20 mm3 and Hamamatsu 8×8 SiPM array with pixel size of 3×3 mm2 and pitch of 3.2 mm was used. The performance of the detector module was measured. All crystals can be clearly resolved from the measured flood histogram. An average DOI resolution of 4.42 mm and an average timing resolution of 433 ps are obtained for events with E>400 keV. The average energy resolution of the detector is 13.1%. The detector module developed in this work can be used to build high performance whole body and total body PET scanners.

Keywords: depth-encoding PET detector; light sharing; single-ended readout; SiPM; row and column readout
Poster panel: 142

Poster Number:

Image Quality Evaluation Study of an RF-Penetrable Brain PET Insert: A Phantom Assessment Toward Clinical Translation (#1632)

A. Groll1, C. - M. Chang1, 2, B. J. Lee1, 3, C. S. Levin1, 4

1 Stanford University , Radiology, Stanford, California, United States of America
2 Stanford, Applied Physics, Stanford, California, United States of America
3 Stanford, Mechanical Engineering, Stanford, California, United States of America
4 Stanford, Electrical Engineering, Stanford, California, United States of America


This work presents imaging studies from a brain dedicated radio frequency (RF)-penetrable PET insert compatible with clinical whole-body MRI systems. The brain dedicated PET system is composed of 16 detector modules. Each module employs an array of 3.2 x 3.2 x 20 mm3 LYSO crystal elements which are 1-1 coupled to arrays of silicon photomultipliers (SiPM). Front-end electronics multiplex the output of 128 pixels to 16 vertical-cavity surface-emitting lasers (VCSEL). The VSCELs generate unique optical output patterns per pixel which are passed via fiber optics to an external DAQ. To achieve RF-penetrability of the body coil RF excitation signal and maintain MRI compatibility, the modules in the assembled system are electrically isolated from the MRI and spaced with 1 mm gaps for the RF fields to enter. The PET system has an internal diameter of 32 cm which is reduced to 28 cm due to the addition of a very thin phased array receive coil. Two phantoms were imaged using the brain dedicated PET system: a custom 3D printed resolution phantom, and the Hoffman brain phantom. The resolution phantom had hot rods with diameters of 5.2 mm, 4.2 mm, 3.2 mm, and 2.8 mm, and a single cold rod region with rods of 4.2 mm diameters. For a gold standard, images from a GE Signa system were acquired and compared to the results of the dedicated brain PET insert. In comparison, the dedicated brain PET system was able to visualize the smallest hot rod feature (2.8 mm) whereas the GE Signa system was not and showed much better resolution and contrast for all rods, including the 4.2 mm diameter cold rod pattern. However, the Hoffman brain phantom scan was of higher quality in the GE system most likely due to the application of accurate image corrections (e.g. random, scatter, and deadtime correction). Future work will focus on the inclusion of image correction in the processing workflow of the dedicated brain PET system to realize the benefit from its higher intrinsic spatial resolution.

Keywords: PET, Clinical, Brain, Phantom
Poster panel: 145

Poster Number:

Optimization of a flat-panel virtual-pinhole PET insert system for improving lesion detectability (#1905)

J. Jiang1, K. Li2, Q. Wang1, D. Tomov1, S. Komarov1, J. A. O'Sullivan2, Y. - C. Tai1

1 Washington University in St. Louis, Department of Radiology, St. Louis, Missouri, United States of America
2 Washington University in St. Louis, Department of Electrical and Systems Engineering, St. Louis, Missouri, United States of America


High resolution detector inserts with virtual pinhole(VP) PET geometry have been demonstrated effective in improving regional spatial resolution and sensitivity of existing clinical and pre-clinical PET scanners. We recently developed a second-generation prototype VP PET insert device that contains SiPM-based PET detectors arranged in a flat-panel geometry. Its panel detector can be placed arbitrarily around a patient’s body using a robotic arm. When working in conjunction with a whole-body PET scanner, the device offers high resolution PET images for any region-of-interest(ROI). New firmware and software were developed to support up to 96 additional detectors of 16×16 crystals each without compromising the scanner functions. This work studied the impact of system design, as well as the placement of flat panel detectors, on contrast recovery of small lesions in order to optimize the performance of the VP-PET insert technology for cancer detection. In the GATE simulation, a total number of 96 high-resolution detectors, each containing 16×16 LYSO crystals, were employed and integrated into a PET scanner. Four different configurations due to different arrangements and placements of the high resolution detectors were simulated. We analyzed the effect of insert crystal thickness and pixel size on the contrast recovery coefficient(CRC) of tumors at diffident ROI by setting the insert crystal thickness to be 3 and 7 mm and the crystal pixel to be 1 and 2 mm. The CRC enhancements of tumors in different ROI under different configurations of insert detectors were also studied. This work reveals that the placement of the high resolution detectors has significant impact on CRC of lesions in different regions. Therefore, the placement of VP-PET insert detectors need to be optimized based on each application. Larger flat-panel insert device will be built and additional imaging studies with different placement of the prototype insert device will also be conducted and presented.

Keywords: VP PET, Optimization, flat-panel insert, GPU reconstruction
Poster panel: 148

Poster Number:

Evaluation of region-of-interest-based brain PET reconstruction (#2198)

J. De Ro1, G. Schramm1, J. Nuyts1

1 KU/UZ Leuven , Department of Imaging and Pathology, Division of Nuclear Medicine, Leuven, Belgium


In standard clinical brain PET imaging, regional quantification is usually
performed after image reconstruction by calculating the mean tracer uptake in the
image in different regions of interest (ROIs) that are defined e.g.
in a high resolution MR image.
Since the achievable reconstructed spatial resolution of MLEM images of
current PET scanners is limited to ca. 4 - 5\,mm,
the calculation of the mean tracer uptake in small regions, such as e.g.
gray matter gyri, is strongly influenced by partial volume effects (PVEs).

To reduce the adverse influence of PVEs, we propose and analyze an approach to directly
reconstruct the mean uptake of a given set of ROIs instead of calculating them in a
post reconstruction step.
In particular, we investigate:
(1) whether the reconstruction-based ROI quantification is superior to the
    post reconstruction quantification.
(2) whether it is better for the quantification of a given ROI, to constrain the
    reconstruction in neighboring pixels (e.g. by also grouping them in ROIs) as
    compared to applying no constraints.
(3) the influence of heterogeneities in a given ROI on the quantification of that ROI
    and its neighboring ROIs.

Reconstructions of 2D simulated PET data from a dedicated brain phantom including 36 ROIs
showed that reconstruction-based ROI quantification
outperformed standard post-reconstruction quantification in terms of bias and noise
even in the presence of local heterogeneities.
Our results indicate that it is better to constrain pixels in regions that are not of interest.
Here this was done by grouping them in two large ROIs, but other noise suppressing constraints are
expected to be effective as well.
In the future, we plan to evaluate the reconstruction-based ROI quantification on
simulated 3D as well as patient data.


Keywords: PET, iterative image reconstruction, quantification
Poster panel: 151

Poster Number:

Multi-Resolution Imaging for Total-body PET (#2406)

W. Huimin1, H. Debin2, Z. Jun1, L. Yang2

1 Shanghai Jiao Tong University, School of Biomedical Engineering, Shanghai, China
2 Shanghai United Imaging Healthcare Co.Ltd, Molecular Imaging, Shanghai, China


Based on the two-meter-long total-body PET scanner, the multi-resolution reconstruction algorithm aims to reconstructing multiple ROI images with different parameters (pixel size, iteration number and post-filter strength, etc) simultaneously. The proposed algorithm is an extension of the conventional LOR-OSEM algorithm. Instead of representing the whole image with only one image matrix, the proposed method uses a set of intermediate matrixes to store the whole image. Each matrix is for one part of the whole image. One LOR may pass through multiple matrixes, and the contributions of voxels in each matrix are calculated individually during the forward and backward projection process. Furthermore, parallel ray-tracing algorithm is developed for fast GPU implementation. The proposed algorithm was validated using a 2D digit phantom and real data collected from a commercial PET/CT scanner with the axial field of view 300mm, both results illustrate significant contrast of reconstructed images with different-resolution parts.

Poster panel: 154

Poster Number:

A Low Profile Imager with Noise-Suppression Readout Based on Gain Stabilized Silicon Photomultipliers (#2521)

J. E. McKisson1, V. Popov1, B. L. Welch2, A. G. Weisenberger1, S. Lee1, J. E. McKisson1, W. Xi1

1 Thomas Jefferson National Accelerator Facility, Newport News, Virginia, United States of America
2 Dilon Diagnostics, Newport News, Virginia, United States of America


We describe the design, construction and initial characterization of an imaging detector module utilizing silicon photomultipliers (SiPM). Our goal is to construct a clinical gamma camera based on arrays of SiPMs coupled to crystal scintillator for Tc-99m imaging for nuclear medicine imaging applications. The design tackles several of the usual problems with implementations using the SiPM. The photodetector elements are mounted in modular units of approximately 51 mm square, each with an 8 x 8 array of photo detectors with independent bias filtering. The overall detector initially contains twelve of these modules, with plans to implement a fifteen-module detector array (150 x 250 mm) in the near future. The initial detector will be SiPM arrays optically coupled  to a segmented NaI scintillator. A passive gain stabilization technique is employed with each photodetector element to both match gains of the SiPMs across the array and to eliminate the gain drift with temperature. Readout of the array is accomplished with an Anger-encoded resistive readout. The challenge of the noise contribution from non-participating rows and columns is addressed with a channel gating suppression method driven by fast logic implemented in a complex programmable logic device (CPLD). Outputs are digitized in a flash ADC.  We describe the initial testing of the detector performance including uniformity, resolution, count rate and performance of the noise suppression.


Keywords: SiPM, SPECT, Tc-99m, Imager, Compact
Poster panel: 157

Poster Number:

Count-rate performance of a PET/CT scanner for 15, 20, and 25 cm axial lengths (#2686)

P. E. Kinahan1, O. Mawlawi4, W. C. J. Hunter1, P. Muzi1, T. Pan4, J. Uribe3, C. W. Stearns3, J. Sunderland2

1 University of Washington, Seattle, Washington, United States of America
2 University of Iowa, Iowa City, Iowa, United States of America
3 GE Healthcare, Waukesha, Wisconsin, United States of America
4 MD Anderson Cancer Center, Houston, Texas, United States of America


The Discovery MI (DMI) PET/CT scanner has a modular design allowing for nominal 15, 20, and 25 cm axial extents (3, 4 or 5 rings). We evaluate the count-rate performance as a function of axial length by pooling data from multiple systems.

The DMI scanner is composed of 34 modules forming a ring with inner diameter of 744 mm. Each module consists of either 3, 4, or 5 detector blocks. In the axial direction, each block assembly contains 9 scintillator crystals that are 5.3 mm wide, resulting in a 48 mm overall width (5 cm nominal). We collected NEMA NU2-2012 count-rate performance data on DMI scanners configured with 3, 4, and 5 rings. The performance was compared at both peak noise equivalent count-rates (peakNEC) and nominal clinical activities (clinicalNEC) defined as the activity concentration typical for a clinical acquisition.

The activity concentration used for clinicalNEC was 3.6 kBq/ml (0.1 uCi/mL). For axial extents of 15, 20, 25 cm, the peakNEC rates were 101, 181, and 269 kcps, and the corresponding clinicalNEC rates were 34, 60, and 96 kcps. The peakNEC and clinicalNEC rates increased with axial length in a manner consistent with previously predicted quadratic behavior. At the clinicalNEC activity concentration, the random coincidence fraction increased from 16 to 18%, while the scattered coincidence fraction was essentially constant at 39%. By using the constraint that the same NEC/mm (axially) is collected for a minimum patient scan length of 790 mm, the scan time per bed position is 240 s, 190 s, and 144 s, for the 3, 4 and 5 ring configurations. Using appropriate bed overlap parameters, the corresponding total acquisitions time are 28 min, 16 min, and 10 min.

The peakNEC and clinicalNEC count-rates increase quadratically with axial extent. From the combination of the increase in NEC and extended axial field of view, each nominal 5 cm increment (i.e. 1 detector ring) decreases total scan acquisition time by approximately 60% for equivalent image SNR.

Poster panel: 160

Poster Number:

Cancer Lesions Detectability Limits in SPECT Breast Imaging (#2395)

I. Lytrosyngounis1, D. Zarketan2, M. Argyrou3, C. Mainta3, D. Maintas3, M. - E. Tomazinaki1, 2, E. Stiliaris1, 4

1 National & Kapodistrian University of Athens, Department of Physics, Athens, Greece
2 National & Kapodistrian University of Athens, Medical School, Athens, Greece
3 Iatrikon Hospital, Institute of Isotopic Studies, Athens, Greece
4 Institute of Accelerating Systems & Applications (IASA), Athens, Greece


Breast imaging for early stage cancer diagnosis is still mainly relying on planar imaging techniques. Due to spatial resolution and sensitivity limitations in combination with absorption effects, whole-body PET and SPECT clinical scanners show a poor performance in three-dimensional tomographic imaging. In order to exercise in detail the detectability limits of small breast cancer lesions in a highly noisy background, a dedicated SPECT study with a breast phantom is presented in this work. The practical goal of the current analysis is to define the tomographic limits of this method and to successfully deliver accurate clinical tomographic images of various small-volume and specific activity concentrations in a controllable Signal-to-Noise environment.

Keywords: Gamma Camera, SPECT, Breast Phantom, GEANT4/GATE, Signal-to-Noise Ratio
Poster panel: 163

Poster Number:

Evaluation of Image Quality and Quantitation in a Clinical PET Scanner with a Uniformly Sparse Detector Rings Configuration (#2807)

N. A. Karakatsanis1, S. Zein1, S. A. Nehmeh1

1 Weill Cornell Medical College, Division of Radiopharmaceutical Sciences, Department of Radiology, New York, New York, United States of America


Positron Emission Tomography (PET) cameras encompass a set of compact crystal detector rings. In this proof of principle study, we introduce a sparse ring configuration involving half the original number of rings uniformly spaced across the same axial Field-of-view (FOV) and assess its effect on image quality and quantitation in PET imaging. The Siemens PET/MR (mMR) system matrix was adopted in our proposed configuration. mMR consists of 64 detector rings made of  4x4x20 mm3 LSO crystals, and extending over 25.6cm axial Field-of-view (FOV). To emulate our sparse rings configuration, counts in sinograms associated with at least one even ring number were zeroed (Sparse-Sinograms). To account for the loss in spatial information, the zeroed sinograms were estimated by linear interpolation in sinogram space (Inter-Sinogram). The PET images for the compact, the sparse and the interpolated sparse sinogram data were reconstructed using the OSEM algorithm (21 subsets, 5 iterations) provided by the Software for Tomographic Image Reconstruction using the original mMR system matrix to maintain the same number of slices in all images. We validated our approach for one brain FDG PET/MR dataset in terms of image quality, target-to-background ratio (TBR) and contrast-to-noise ratio (CNR) scores for different number of OSEM iterations. Sparse rings configuration yielded comparable image quality to that of the clinical dataset. TBR and CNR showed increased error with the number of OSEM iterations (8.3% and 23.6% respectively at 5 iterations), decreasing to 1.7% and 5.4% respectively in the Inter-Sinogram images. PET imaging with half the number of original detector rings uniformly spaced over the same axial FOV, yielded comparable image quality, yet reduced TBR and CNR, which may be recovered via linear inter-sinogram interpolation. Uniformly spacing a given set of PET detector rings to double their axial FOV is possible without significant losses in image quality and quantitation.

Keywords: PET, sparse, ring, configuration, sinogram interpolation
Poster panel: 166

Poster Number:

Comparison of flat field correction methods for photon-counting spectral CT images (#1326)

D. Kim1, 2, J. Baek1

1 Yonsei University, School of Integrated Technology, Incheon, Republic of Korea
2 Samsung Electronics, DR Development Group, Suwon, Republic of Korea


For the practical usage of the photon-counting detector (PCD) in medical imaging, conductor material stability, photon counting efficiency, and energy separation accuracy must reach a certain level. Besides the pulse pile-up phenomenon, PCD has a non-linear response of the incident X-ray energies, and thus without proper energy calibration, the reconstructed images produce significant ring artifacts. In this study, we propose a water gain correction method for energy calibration of the PCD. To calculate the gain of each detector pixel, we measured the ratio of the sinogram between a water cylinder phantom and an ideal water cylinder phantom, and then averaged the ratio over all views. Then, this gain was multiplied to the projection data acquired from the traditional flat field correction method. The performance of the proposed method was compared to flat field correction methods that use averaged air shots and scan data of a water cylinder phantom. Our results show that the proposed method reduces the ring artifacts effectively without increasing the noise in the final image. It is also observed that the proposed method provides improved SNR and CNR by 45 ~ 55 % compared to the flat field correction using scan data of a water cylinder phantom.

Poster panel: 169

Poster Number:

Visualizing of Breast Blood Vessels without Contrast Medium via a Grating Interferometer with a Conventional X-ray Tube (#1588)

X. Li1, 2, Z. Chen1, 2, L. Zhang1, 2, X. Zhu1, 2, S. Wang3, 4, W. Peng3, 4

1 Ministry of Education, Key Laboratory of Particle & Radiation Imaging, Beijing, China
2 Tsinghua University, Department of Engineering Physics, Beijing, China
3 Fudan University Shanghai Cancer Center, Department of Diagnostic Radiology, Shanghai, China
4 Fudan University, Department of Oncology, Shanghai Medical College, Shanghai, China


Objectives Grating-based X-ray phase-contrast imaging (GPCI) has received growing interests in recent years thanks to its high capability of visualizing soft tissue. Breast phase-contrast imaging is drawing great attentions as it is one of the most promising candidates for clinical applications of this modality. In this work, visualization of breast blood vessels via GPCI computed tomography (CT) is investigated with a laboratory X-ray tube and without any contrast medium.

Methods A deconvolution method was applied to retrieve attenuation, phase-contrast and dark-field images from the phase-stepping curves. In order to overcome the limitation that blood vessels in CT images are too tiny and vimineous to be observed in a single slice, minimum intensity along the vertical direction were utilized on attenuation and phase-contrast images, while maximum intensity along the vertical direction on dark-field images. In addition segmentation was performed to reveal the formation of each vascular structure on dark-field images.

Results For the three types of images retrieved with the deconvolution method, it’s found that blood vessels are obviously visible on all of them but dark-field images is capable of providing more detailed fine structures than the others. By using segmentation, the formation of each vascular structure are clearly visible on dark-field images.

Conclusions It’s concluded that even with a conventional X-ray tube and without any contrast medium, GPCI is of high potential to improve the diagnosis of breast cancer.

Keywords: breast, phase-contrast CT, blood vessels
Poster panel: 172

Poster Number:

Effective Atomic Number and Electron Density Estimation Using Current-Mode Energy-Resolved Computed Tomography (#1714)

T. Hamaguchi1, I. Kanno1

1 Kyoto University, Nuclear Engineering, Kyoto, Japan


In this paper, the estimation method of the effective atomic number Zeff and the electron density Ne using energy-resolved computed tomography (CT) with a transXend detector is described. The transXend detector consists of several segmented detectors, measures X-ray photons as electric currents and estimates incident X-ray spectrum by spectrum unfolding analysis. Using the transXend detector, we can obtain the linear attenuation coefficient distribution image of a phantom as a function of X-ray energy. Zeff and Ne are estimated by comparing the linear attenuation coefficients with the lookup table on photon-atom cross sections given by National Institute of Standard and Technology, USA. We performed energy-resolved CT measurement on the 3 cm diameter cylindrical PMMA phantoms with resin rods, which covered atomic number from 5.5 to 14 and electron density from 3 to 6 ×1023 cm-3. The measured Zeff and Ne showed excellent agreements with the calculated ones.

Keywords: Computed Tomography, Energy-resolved CT, Effective Atomic Number, Electron Density
Poster panel: 175

Poster Number:

System Optimization for Single-grid Phase-contrast X-ray Imaging (#1802)

H. Lee1, H. Lim1, D. Jeon1, H. Cho1

1 Yonsei University, Radiation convergence engineering, Wonju, Republic of Korea


The phase images retrieved by Fourier domain analysis used in a single-grid phase-contrast X-ray imaging (PCXI) are often distorted by image artifacts such as moiré and wraparound. In this study, we propose a modification to a single-grid PCXI system using a Fourier domain analysis technique to extract absorption, scattering, and differential phase-contrast images. The proposed modification is to rotate the X-ray grid in the image plane to achieve spectral separation between the desired information and the moiré artifact, which is introduced by the superposition of the periodic image of the grid shadow and the periodic sampling by the detector. In addition, we performed some system optimization by adjusting distances between source, object, grid, and detector, and grid strip density projected on the detector plane to further improve image quality. The table-top setup used in the experiment consisted of a focused-linear grid with a 200-lines/inch strip density, a microfocus X-ray tube with a 35-μm focal spot size, and a CMOS flat-panel detector with a 49.5-μm pixel size. The X-ray grid was rotated at 76.5 degree with respect to the detector and the sample was placed as close as possible to the X-ray tube. Our results indicated that the PCXI artifacts were effectively eliminated from extracted images, thus improving image quality.

Keywords: Phase-contrast X-ray imaging, Wraparound artifact, Moiré artifact, System optimization
Poster panel: 178

Poster Number:

Real-Time Patient-Specific CT Dose Estimation using a Deep Convolutional Neural Network (#1974)

J. Maier1, 2, E. Eulig1, 2, S. Dorn1, 3, S. Sawall1, 3, M. Kachelrieß1, 3

1 German Cancer Research Center (DKFZ), Division of X-Ray Imaging and CT, Heidelberg, Baden-Württemberg, Germany
2 University of Heidelberg, Department of Physics and Astronomy, Heidelberg, Baden-Württemberg, Germany
3 University of Heidelberg, Medical Faculty, Heidelberg, Baden-Württemberg, Germany


Due to the potential risk of ionizing radiation, the assessment of the administered radiation dose is an important topic in CT. However, dosimetric quantities that are routinely evaluated in CT only refer to the absorbed dose within cylindrical phantoms and do not appropriately represent the actual patient dose. While Monte Carlo (MC), the gold standard for patient-specific dose estimation, is too slow to be applied routinely, faster alternatives are usually far less accurate. To overcome this drawback, we developed the deep dose estimation (DDE) algorithm. DDE uses a deep convolutional neural network to reproduce MC dose estimates given only a CT image and a first-order dose estimate as two-channel input. To learn the corresponding mapping, DDE was trained on artificial data generated from whole-body clinical CT scans of 15 patients. For each patient 60 circular CT scans were simulated for 20 different z-positions (pelvis, abdomen, thorax) and 3 different acquisition settings. The total number of 900 data sets was divided into 720 training data sets (12 patients) and 180 validation data sets (3 patients). Each scan was reconstructed on a 256 × 256 × 48 voxel grid with an isotropic voxel size of 2 mm. In addition a first-order dose estimate as well as a MC dose estimate was calculated for every data set. Using these data, the network's open parameters were determined by minimizing the mean relative error (MRE) between the output and the MC dose estimate. Evaluating the MRE on the validation data set yields deviations of 3.0 % on average with respect to the ground truth and processing times of 250 ms for a CT scan with 10 cm longitudinal coverage. Thus, DDE is able to achieve similar accuracy as MC while performing orders of magnitude faster.

Keywords: Patient-specific dose estimation, Deep Learning, Computed tomography, Deep convolutional neural network, Monte Carlo
Poster panel: 181

Poster Number:

An interior Compton Scatter Tomography  (#2055)

J. Cebeiro1, M. K. Nguyen2, M. A. Morvidone1, C. Tarpau2

1 Centro de Matemática Aplicada, Universidad Nacional de San Martín, Gral. San Martín, Argentina
2 Equipes de Traitement de l'Information et Systèmes (ETIS), Université de Cergy-Pontoise/ENSEA/UMR CNRS 8051, Cergy-Pontoise, France


A new modality of Compton Scatter Tomography (CST) for interior problem is presented. As in the original modality introduced by Norton, it has a fixed source but the detector site is no longer on a line passing through the source but on a circle on which the source is located. This modality is more adapted to small objects whereas the Norton's modality is conceived for large objects and one side scanning. The solution of the corresponding Radon problem is shown to be obtained by appropriate geometric inversion of known Radon problems. We show numerical simulations illustrating this idea.

Keywords: transmission imaging, interior problem, Compton Scatter Tomography, Tomographic reconstruction methods
Poster panel: 184

Poster Number:

Cone-Beam CT for Breast Specimens in Surgery: The Phantom Study (#2289)

S. S. Thongvigitmanee1, S. Aootaphao1, C. Thanasupsombat1, A. Kiang-ia2, W. Narkbuakaew1, K. Wangkaoom1, P. Junhunee2, S. Laohawiriyakamol3, P. Puttawibul3, P. Thajchayapong4

1 National Electronics and Computer Technology Center, X-Ray CT and Medical Imaging Lab, Pathumthani, Thailand
2 National Metal and Materials Technology Center, Medical Devices Lab, Pathumthani, Thailand
3 Prince of Songkla University, Faculty of Medicine, Songkla, Thailand
4 National Science and Technology Development Agency, Pathumthani, Thailand


During breast surgery, an immediate verification method of the negative resection margin in a breast cancer specimen is needed. Conventionally, a mammogram of a specimen has been used; however, it is in 2D and the resection margin of abnormal lesion and mass can be suspicious. The aim of this paper is to propose the cone-beam CT machine specifically for breast specimens to verify the tumor free margin during breast surgery and evaluate its performance. All setting parameters were designed not only to detect lesion inside breast specimens, but also to perform fast acquisition and 3D reconstruction. Image quality assessment was evaluated using standard and simulated phantoms. The spatial resolution was found at 2.8 line pair per mm, while the cupping artifacts and noise were about 1.2% and 2.9%, respectively. Moreover, calcium carbonate in a rasin phantom representing microcalcification in a breast specimen can be detected with this machine. Thus, the proposed CBCT scanner showed high potential to detect microcalcification within breast specimens during surgery.

Keywords: Cone-beam CT, Breast specimen, Image Reconstruction
Poster panel: 187

Poster Number:

Simulation of Dose Rate and Mass Distribution of Gamma CT (#2421)

H. Matsunaga1, H. Kawai1, A. Kobayashi1

1 Chiba University, Graduate School of Science and Engineering, Chiba, Japan


X-CT is widely spread to see through the body at the medical site. In the measurement of X-CT at once, the patient get medical exposure about 5-30mSv. Decreasing amount of exposure is one of the task in the medical site. We propose new CT using gamma ray which is available to decrease amount of medical exposure. This CT use gamma ray emitted from electron positron pair annihilation. The pair annihilation is caused by radiating β+ rays emitted from positron emitting nuclides to inorganic scintillator. The mass distribution in the body is estimated from transmittance of the gamma ray to the body and the scattering position of single Compton scattering. We make a distribution of single Compton scattering positions in the body using Geant4 and simulate that it corresponds to mass distribution. And we simulate the dose rate of measurement of Gamma CT by Geant4.

Keywords: Gamma CT, dose rate, mass distribution, new CT
Poster panel: 190

Poster Number:

Synchrotron X-ray phase-contrast micro-CT to morphometric study of the Thoropa miliaris tadpole (#2612)

G. Fidalgo1, M. V. Colaço2, L. P. Nogueira2, 4, D. Braz1, H. R. Silva3, G. Colaço3, G. Sena1, A. Pickler1, R. C. Barroso2

1 Federal University of Rio de Janeiro, Nuclear Engineering Program, Rio de Janeiro, Brazil
2 State University of Rio de Janeiro, Physics Institute, Rio de Janeiro, Brazil
3 Federal Rural University of Rio de Janeiro, Biology Institute, Rio de Janeiro, Brazil
4 University of Oslo, Oral Research Laboratory / Institute of Clinical Dentistry, Oslo, Norway


Micro-CT is the most powerful 3D imaging technique to study morphology of the small animals, such as annelids, insects and amphibian embryos. Commercial lab-based X-ray micro-CT system is the most used to study anurans, however it presents some limitation when compared with synchrotron micro-CT techniques that offers significant advantages over such as a wide energy range, high flux and small source size and beam divergence. In this work, we performed the first morphometric study of internal structures of Thoropa miliaris tadpoles, using synchrotron phase-contrast micro-CT.
A tadpole is the free-living aquatic larva of an anuran. As a larva, it is non-reproductive and undergoes radical changes in morphology and ecology between life stages. About three quarters of 7,537 species have a tadpole for some period during their development, and tadpole descriptions are available for about a third of those. Most of tadpole structure is composed by low attenuation tissue, becoming an interesting specimen for phase-contrast micro-CT analysis.
High-resolution phase-contrast micro-CT scans were performed at the SYRMEP beamline. In this set-up, the out coming beam from the ring is restrained before the monochromator and in a devoted end-station, absorption and phase contrast radiography and tomography set-up are available. The phase-contrast effect provides appropriate edge enhancement of tadpole structures. It was possible to distinguish structures to apply virtual segmentation process and volume quantification for different organs.

Keywords: microtomography, micro-CT, phase-contrast, virtual dissection, Thoropa miliaris
Poster panel: 193

Poster Number:

Investigation of Photon Counting CdZnTe Detectors Under High Flux X-ray Irradiation (#2837)

Y. Li1, S. Xi1, Z. Du1, P. Zhu1, G. Zha1, W. Jie1, J. Dong1

1 Northwestern Polytechnical University, School of Materials and Science, Xi'an, China


CdZnTe as a kind of excellent room temperature radiation detection materials is widely used in X-ray and Gamma-ray detection, In this paper, the performance of photon counting 2D-array CdZnTe detectors under high flux x-ray irradiation was studied, The counting rate value and variation with the x-ray flux change has been investigated, The thermal simulation current (TSC) was used to characterize the point defects levels of different CdZnTe detectors. High counting detectors have low concentration of point defect density ,both shallow level and deep level defects, while the low counting detectors have high concentration of point defects, The X-ray photocurrent curve of CdZnTe detectors is tested, high counting detectors has clear step when changing the flux and the current are stable, for the low counting detectors, the X-ray photocurrent curve does not have clear step when changing the flux, obvious shake occurs when the flux increase to a certain value. A multi-energy X-ray scanning imaging system was built using the detectors with good performance and the imaging is carried out.


Keywords: CZT, photon counting, TSC, imaging
Poster panel: 196

Poster Number:

Measurements of MR-compatibility of LGSO crystals with different Gd concentrations (#1555)

K. Nakanishi1, C. Komoda1, S. Koyama1, Y. Anzai2, J. Osada2, K. Nagao2, H. Ishibashi2, T. Kuriwada3, K. Hazu3, S. Yamamoto1

1 Nagoya University, Graduate School of Medicine, Nagoya, Japan
2 Oxide, Yamanashi, Japan
3 Mitsubishi Chemical Corporation, Tokyo, Japan


It is commonly believed that scintillators containing gadolinium (Gd) produce artifacts in MR-images. However the relation of the artifact levels and the Gd concentration is not obvious. Thus we measured the MR-images of a phantom with LGSOs of different Gd concentrations. LGSO crystals (4 mm x 4 mm x 12 mm) with different Gd concentrations (0 to 2.68 weight %) were arranged on a plastic phantom containing NiSO2 solution (11 cm diameter and 20cm long), MR-images were acquired with various sequences of 3T MRI system (Siemens MAGNERTOM Verio 3T). For all images of different sequences of the MRI, image artifacts were observed for the LGSO with Gd concentrations higher than 0.58 weight %. Below this concentrations, no difference of the images was observed. We conclude that below 0.58 weight %, LGSO will not produce artifacts in MR images for any pulse sequences.

Keywords: PET/MRI, LGSO
Poster panel: 199

Poster Number:

A Novel Theranostic Nano-Platform For Simultaneous Multimodal Imaging And Radionuclide Therapy (#2038)

Y. H. Gholami1, 2, M. Q. Wilks3, R. Maschmeyer1, H. Yuan3, M. Normandin3, L. Josephson3, G. El Fakhri3, Z. Kuncic1

1 The University of Sydney , School of Physics, Institute of Medical Physics, Sydney , Australia
2 Sydney Vital Translational Cancer Research , Kolling Institute of Medical Research, Sydney, Australia
3 Harvard Medical School, Massachusetts General Hospital, Gordon Centre for Medical Imaging, Boston, Massachusetts, United States of America


A theranostic nanoparticle platform has been developed to enable simultaneous multimodal imaging and radionuclide therapy to improve detection, diagnosis and targeted treatment of metastases in the lymph nodes. The platform is based on a Super-Paramagnetic Iron Oxide Nanoparticle (SPION) that enhances contrast in Magnetic Resonance Imaging (MRI) and also enables simultaneous Positron Emission Tomography (PET) upon radiolabeling with a PET tracer. Integrating the high spatial resolution of MRI with the high sensitivity of PET affords a significant image quality gain. The SPION-based nanoparticle Feraheme was used as the nanoparticle platform as it is FDA approved and is increasingly being used off-label as an MRI contrast agent for diagnostic lymph node imaging. Following intravenous injection, Feraheme nanoparticles slowly extravasate from vascular space to interstitial space, from where they can be taken up by immune cells and delivered via lymphatic vessels to lymph nodes. The nanoparticles remain in normal nodal tissue and reduce MRI signal intensity, thereby enhancing contrast against any metastatic lesions in the node. Feraheme was radiolabeled with the PET radioisotope 89Zr using a novel chelate-free radiolabeling technique, heat induced radiolabeling. The radiolabelling technique was further adapted for 177Lu and 90Y radiotherapeutic labelling of Feraheme. Radiochemical analysis demonstrated a high radiochemical yield (up to ≈ 96%) and purity ( up to ≈ 98%) of the radiolabeled Feraheme products. Pre-clinical simultaneous PET-MRI scans confirmed the capability of the 89Zr-labelled Feraheme nano-platform for this multi-modal imaging technique.

Keywords: PET/MRI, radiolabelling, theranostic, nanomedicine
Poster panel: 202

Poster Number:

Microstrip transmission-line array coil dedicated for PET insert for MRI system (#2242)

M. S. H. Akram1, T. Obata1, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology (QST), National Institute of Radiological Sciences (NIRS), Chiba, Japan


At ultrahigh magnetic field, like 7 Tesla, the nuclear magnetic resonance (NMR) frequency is high enough (e.g., 300 MHz) to make the radiofrequency (RF) wavelength become comparable to the imaging dimension of the human organs which generates constructive and destructive interferences of the field in the imaging region (ROI). The conventional transmit volume RF coil, like the birdcage coil, is challenging to implement with high RF field homogeneity for ultrahigh field MRI study. Multichannel RF coil, like microstrip transmission line array coil, is a potential alternative, in which case the phase and magnitude of the RF pulse in each channel is possible to manipulate to generate homogeneous RF field in the required imaging region.

A microstrip transmission line coil consists of a microstrip conductor connected to the RF feed line, a ground conductor plane that reflects the field created the by the microstrip conductor and, dielectric materials in between them. On the other hand, the PET detector modules for the MRI system are RF shielded to avoid mutual interference between PET and MRI systems. In this study, the RF shield of the PET insert is implemented as the ground plane for the microstrip transmission line coil. We developed a concept design of this coil for our available 3 Tesla MRI system. Since our existing 3 T MRI system has single RF transmission channel connected to the MRI patient bed, we developed a four channel RF power dividing interface circuit for feeding the four channel RF coil. The coils were implemented both as transmitter and receiver. Both the reflection and transmission coefficients of S parameters were above -19 dB with VSWR close to 1. The S parameters were measured without using any decoupling circuits between coils and mounting on a cylindrical frame keeping wider gap in between two coils. The homogeneous phantom studies show SNR of around 800 and RF field homogeneity of 88% for the central 55-mm diameter imaging region. 

Poster panel: 205

Poster Number:

MR-compatibility Assessment of PETsys TOF ASIC (#2306)

Y. T. Kim1, S. Kim2, J. - M. Kim3, M. N. Ullah1, J. Park1, C. - H. Oh3, J. - Y. Yeom2, 1

1 Korea University, Department of Bio-convergence engineering, Seoul, Republic of Korea
2 Korea University, Department of Biomedical engineering, Seoul, Republic of Korea
3 Korea University, Department of Electronics & Information Engineering, Sejong-si, Republic of Korea


PET/MRI (Positron Emission Tomography / Magnetic Resolution Imaging) is a promising hybrid modality in the medical imaging field. However, PET-MRI mutual compatibility issues have always been challenging when integrating the two modalities. In this study, we tested the MR compatibility of a commercial time-of-flight (TOF) capable ASIC from PETsys Electronics. The performance of the PETsys TOF ASIC evaluation kit was tested in a Philips Achieva 3.0T TX scanner using custom 3 m long cables that connect the front-end
(including KETEK PM3325-WB SiPM and 3 x 3 x 5 mm3 LYSO crystals) to the dedicated data acquisition system. Measurements were performed in the MRI with 1) B0 only, 2) TSE (Turbo Spin Echo), 3) FFE (Fast Field Echo) sequences, and 4) the MRI device off with a 22Na source. The ASIC chip operated well in the 3T MRI scanner, but slight timing and energy performance degradation was observed for TSE sequence. More detailed studies are scheduled including better temperature controls, and additional data (MRI images, etc.) will be provided.

Poster panel: 208

Poster Number:

Towards MRI compatibility of a PET detector based on the HPK C13500 series PET module (#2411)

F. P. Schmidt1, C. Parl1, R. Ladebeck2, J. Breuer3, N. Zhang4, M. Schmand4, B. J. Pichler1

1 University of Tuebingen, Werner Siemens Imaging Center, Department of Preclinical Imaging and Radiopharmacy, Tuebingen, Baden-Württemberg, Germany
2 Siemens Healthcare GmbH, Diagnostic Imaging, Magnetic Resonance, Erlangen, Bavaria, Germany
3 Siemens Healthcare GmbH, Diagnostic Imaging, Molecular Imaging, Forchheim, Bavaria, Germany
4 Siemens Medical Solutions USA Inc., Diagnostic Imaging, Molecular Imaging, Rockford, Tennessee, United States of America


Our project goal is to foster the next level of multimodality imaging by the combination of positron emission tomography and magnetic resonance imaging (PET/MRI). Hereby a crucial part is to ensure that the mutual interference between the PET detector and the MR system is kept at a minimum. A modified version of the C13500-4075LC-12 PET module (Hamamatsu Photonics, Hamamatsu, Japan), the EVAKIT01, is used in this study as it features a data communication based on optical fiber. To test the MRI compatibility, measurements with a 2 mm aluminum alloy shielding inside a clinical 3T MR scanner are compared to a laboratory reference measurement. Radio frequency (RF) sequences with high peak power (flip angle (α)=720 °, repetition time (TR)=104 ms) and high average power (α=62 °, TR=3 ms) and gradient switching sequences (20 mT/m, 0.3 ms ramp) are utilized to stress the operation of the PET module. The average FWHM energy resolution over all application specific integrated circuit (ASIC) channels with 11.9± 0.6% for the MRI sequences is comparable to the reference measurement. This also holds true for the singles count rate of 102.0 kcounts/s and a count distribution of 14.2±3.1 kcounts/s amongst the ASIC channels. An investigation of the 15 ps resolution time stamp register revealed no behavior that indicates distortion due to the MRI sequences. The RF noise spectrum, with a bandwidth of 1.0 MHz around the Larmor frequency, while the EVAKIT01 is operated shows no change in the noise floor or amplitude in comparison to a reference spectrum acquired without the PET module installed. This study shows, that the EVAKIT01 PET module, if properly shielded, has no performance degradation due to the presence of MRI sequences and no RF distortion of the MR system is observable. Measurements with a second PET module to acquire time relevant coincident data and a custom-made 35 µm copper shielding to enable clinical relevant imaging sequences are in preparation.

Keywords: PET, MRI compatibility, ASIC, MR
Poster panel: 211

Poster Number:

Investigation of Dual-Head PET for Dedicated PET/CT Breast Imaging in a Spherical-Segment Geometry (#2760)

A. Z. Register1, T. G. Turkington2, 1, M. P. Tornai2, 1, S. J. Lee3, B. Kross3, A. G. Weisenberger3

1 Duke University, Biomedical Engineering, Durham, North Carolina, United States of America
2 Duke University Medical Center, Radiology & Medical Physics, Durham, North Carolina, United States of America
3 Thomas Jefferson National Accelerator Facility, Detector Group, Neport News, Virginia, United States of America


We investigate large area parallel plate Positron Emission Mammography detectors for incorporation into a versatile 3D positioning gantry for use in high-performance breast PET along with x-ray Computed Tomography (PET/CT). This dedicated hybrid imaging system will interrogate breast health. We modify an existing hybrid system capable of traversing complex arbitrary 3D spherical-segment trajectories. This set up ensures more nearly-complete sampling throughout the common pendant-breast imaging volume compared with other dual-detector breast PET geometries. Each detector is composed of 15 x 20cm arrays of LGSO crystals from an existing PEM system. Breast compression is not needed. Asymmetric detector positioning allows system tilting such that the external lateral detector can be posterior to the breast and in coincidence with a more proximal detector anterior-obliquely located near the breast. In such an orientation, the anterior chest wall (and potentially the axilla) can be imaged with the coincident detector pair. A diagonally oriented line source in a 7cm diameter cylinder with and without water is rotated between the detectors at different separation distances; similar studies are performed with 3 isolated spherical lesions. Various other configurations are tested including detector system tilting to obtain a spherical-segment (banded) acquisition. Initial results indicate that the detector distance in the asymmetric mode has a greater effect on measured background signal level than the stochastic gamma emissions from the phantoms. There is contamination from the PEM-like imaging nature of the system in its current configuration. Improvements are ongoing.

Keywords: PET, CT, Breast Imaging, PET/CT, Dedicated System
Poster panel: 214

Poster Number:

Low Dose and Sparse View CT Reconstruction via Generative Adversarial Networks (#1206)

Y. Sun1, Z. Zhao1, P. Cong1

1 Tsinghua University, Institute of Nuclear and New Energy Technology, Beijing, China


Low dose and sparse view CT are effective approaches to reduce the radiation dose and accelerate scan speed. Images reconstructed from insufficient data acquired from low dose and sparse view CT are associated with severe streaking artifacts. Therefore, reducing the radiation dose will further degrade the imaging quality. Several attempts have been made to remove these artifacts using deep learning methods such as CNN. Although the deep learning methods for low dose and sparse view CT reconstruction have gained impressive successes, the reconstruction results are still over-smooth. In this work, we propose an artifacts reduction method for low dose and sparse-view CT via a single model trained by generative adversarial networks (GAN). Several numerical simulation experiments are implemented to test the performance of our network. The results show that our GAN can significantly reduce the streaking atrifacts campared with FBP mehtod, and carries more detailed information than CNN.

Keywords: low dose, sparse view, GAN
Poster panel: 217

Poster Number:

State estimation in dynamic MRI (#1299)

V. - V. Wettenhovi1, V. Kolehmainen1, J. Huttunen2, M. Kettunen2, O. Gröhn2, M. Vauhkonen1

1 University of Eastern Finland, Department of Applied Physics, Kuopio, Finland
2 University of Eastern Finland, A.I. Virtanen Institute for Molecular Sciences, Kuopio, Finland


We propose the use of a state estimation method in dynamic magnetic resonance imaging (MRI) reconstruction. In state estimation, we model the time-dependent system by using separate models for the state evolution and observations. In our method, the state estimation model is enhanced with a spatial prior that contains information of the estimate in the form of an anatomical prior. This spatial prior promotes structural similarity of the gradient of the unknown image with the gradient of the anatomical image. The spatial prior is embedded into the state evolution model, leading to a computationally less intensive formulation compared to our previous approach, where the observation model was augmented with the spatial prior. In this work, the Kalman filter (KF) is used to compute the state estimates. This formulation is evaluated by using experimental dynamic contrast-enhanced MRI data from a small animal specimen and simulated functional MRI data, both acquired using radial sampling with golden angle acquisition. The results are compared to conventional sliding window method and the previously used augmented KF.

Keywords: Kalman filter, Magnetic resonance imaging, GPU, radial MRI
Poster panel: 220

Poster Number:

A Dual-GPU Approach to 3D List-mode PET Image Reconstruction (#1354)

M. Teimoorisichani1, A. L. Goertzen1, 2

1 University of Manitoba, Physics and Astronomy, Winnipeg, Manitoba, Canada
2 University of Manitoba, Radiology, Winnipeg, Manitoba, Canada


Graphics processing units (GPUs) have been employed to accelerate PET image reconstruction over the past years. GPU-based techniques can be of particular interest when list-mode reconstructions are desired. In this work, we propose a dual-GPU image reconstruction that splits the reconstruction tasks to both GPUs in parallel. The two distinct features of the proposed dual-GPU approach are 1) accelerating the reconstruction by splitting both the forward and back-projection steps to both devices, and 2) supporting a larger data size when existing data are too large to fit in the memory of a single GPU. The proposed list-mode reconstruction is based on a tube-of-response (TOR) calculation for each recorded coincidence. Elements of the system matrix for each event (voxel indices and their contribution coefficients within a TOR) are calculated on-the-fly on GPU. We proposed to split the reconstruction volume equally over the GPUs and process each image segment using a tiling technique. The forward projection values obtained from all GPUs can be merged using a high throughput peer-to-peer (P2P) access between the GPUs. Also, the use of CUDA streams are explored in forward projection. The proposed method was evaluated on a dual-GPU setup where the two GPUs are connected over the PCIe bus of a single node. The list-mode image reconstruction technique was seen to accelerate up to twice when running on two GPUs rather than a single GPU and be tens to hundreds of times faster (depending on reconstruction parameters) than an equivalent single-threaded CPU reconstruction.

Keywords: Iterative image reconstruction, GPU
Poster panel: 223

Poster Number:

ConCerto: Generic Library for CT Pseudo Monte Carlo Simulation and Reconstruction (#1442)

D. Benoit1, M. Sibomana2, D. Visvikis1

1 Laboratory of Medical Information Processing (LaTIM, INSERM UMR1101), CHRU Brest, Brest, France
2 Université catholique de Louvain, ICTEAM/ELEN, Louvain-La-Neuve, Belgium


CT (including CBCT) data simulation using Monte Carlo is very time consuming. Similarly, iterative reconstruction dedicated to CT is slow compared to classical analytical reconstruction such as FDK. Given the improvements in computational hardware domain (GPU, XeonPhi), new and faster software can be implemented. The Monte Carlo simulation platforms GATE (CPU only) and GGEMS (GPU) are classical ways to simulate very realistic data for CT. On the other hand, image reconstruction platforms such as RTK (dedicated to CBCT reconstruction) or CASToR (dedicated to generic CT, using only CPU) for CT data reconstruction. In this context, a new library named ConCerto is proposed providing very fast pseudo MC simulated CT data for the validation of CT/CBCT image reconstruction algorithms, some of which are included in this library. The proposed development is a library that can be used across different computer hardware platforms using CPUs and/or GPUs. For this reason the OpenCL library is chosen. Furthermore the library is supported on all main operating systems: Windows, macOS or Linux. Using the ConCerto simulation tool reduces the computation time by a factor 93 on 3 GPUs compared to one CPU (1 thread) for a voxellized phantom of 184x103x161 voxels. Iterative CT reconstruction using ConCerto and 3 GPUs is also the fastest configuration reducing computation time by a factor 3.61 compared to using CPUs (4 cores / 8 threads). Simulated results show that estimated attenuation values in all regions of interest are very close to the reference with an error inferior to 1%. In future, other reconstruction algorithms such as MLTRC, FDK or IMPACT will be implemented, including alternative projectors such as distance driven. Finally ConCerto will be generalised to perform spectral CT and Dual CT reconstruction.

Keywords: Monte Carlo Simulation, Reconstruction, CT, GPU
Poster panel: 226

Poster Number:

Dynamic PET Reconstruction Utilizing a Spatiotemporal 4D De-noising Kernel  (#1626)

J. - C. Cheng1, 2, C. W. J. Bevington2, A. Rahmim3, I. S. Klyuzhin4, J. Matthews5, R. Boellaard6, 7, V. Sossi2

1 The University of British Columbia, Pacific Parkinson's Research Centre, Vancouver, British Columbia, Canada
2 The University of British Columbia, Department of Physics and Astronomy, Vancouver, British Columbia, Canada
3 Johns Hopkins University, Department of Radiology and Radiological Sciences, School of Medicine, Baltimore, Maryland, United States of America
4 The University of British Columbia, Department of Medicine, Division of Neurology, Vancouver, British Columbia, Canada
5 The University of Manchester, Division of Informatics, Imaging and Data Sciences, Manchester, United Kingdom
6 VU University Medical Center, Department of Radiology and Nuclear Medicine, Amsterdam, Netherlands
7 University Medical Center Groningen, Department of Nuclear Medicine and Molecular Imaging, Groningen, Netherlands


We propose novel 4D de-noised image reconstruction frame works, followed by extensive validation using 4D simulations, experimental phantom as well as clinical patient data. Previously, it was demonstrated that our 3D de-noised reconstruction, which applies the HighlY constrained backPRojection (HYPR) de-noising operator After each Update of OSEM (HYPR-AU-OSEM), can achieve noise reduction and improve the reproducibility in contrast recovery without degrading accuracy in terms of resolution and contrast for single frame reconstruction. Moreover, the method does not require any prior information and is not computationally intensive. In this work, we propose the 4D extension of HYPR-AU-OSEM (i.e. HYPR4D-AU-OSEM) for dynamic imaging. Further, we incorporate the proposed 4D de-noising operator within the recently proposed machine learning based kernelized reconstruction frame work (i.e. HYPR4D-K-OSEM). In short, the proposed methods make use of the spatiotemporal high frequency features extracted from the 4D composite, generated directly within the reconstruction, to preserve the 4D resolution and constrain the noise increment in both spatial and temporal domains. Results from the simulations, experimental phantom, and patient data showed that the proposed methods outperformed the standard OSEM with post filter in terms of 4D resolution, contrast recovery coefficient vs noise trade-off, and accuracy in time-activity-curves (TAC) and binding potential (BPND) values. In particular, the underestimation in regional BPND values was reduced from ~27% to ~7% using the proposed methods. Compared to the conventional 3D composite, the 4D composite achieved 50% lower mean absolute error in TACs. Comparable results were obtained between AU and kernel methods. In summary, the improvement in 4D resolution and noise reduction obtained from the proposed methods can produce more robust and accurate image features without any prior information, as compared to the conventional methods.

Keywords: 4D de-noising, Dynamic PET reconstruction, Kernel method
Poster panel: 229

Poster Number:

Effect of Bayesian-penalized Likelihood Reconstruction in O-15 H2O Myocardial Blood Flow in Rest and Stress Studies (#1687)

J. Teuho1, 2, C. Han1, V. Saunavaara1, 2, T. Tolvanen1, 2, R. Siekkinen1, 3, M. Teräs2, 4

1 Turku PET Centre, University of Turku and Turku University Hospital, Turku, Finland
2 Department of Medical Physics, Division of Medical Imaging, Turku University Hospital, Turku, Finland
3 Department of Physics and Astronomy, University of Turku, Turku, Finland
4 Department of Biomedicine, University of Turku, Turku, Finland


Introduction: The Discovery MI digital PET/CT system (GE Healthcare) implements a Bayesian-penalized likelihood (BPL) reconstruction technique (Q.Clear), offering image quality improvements over traditional iterative reconstruction techniques (VUE Point FX-S). We studied the effect of BPL in quantification of myocardial blood flow (MBF) in rest and stress perfusion studies with O15-H2O.
Materials and Methods: Three stress studies and one rest study reconstructed with both VUE Point FX-S and Q.Clear were compared. Analysis and kinetic modelling of MBF were performed in Carimas 2.9. We compared the MBF from 4- and 17-segment polar map in both reconstructions. Relative differences and absolute differences for MBF between the algorithms were calculated.
Results: 17-segment polar maps showed only small differences between the reconstruction algorithms. In majority of the polar map regions, the maximum differences were between -5 % to + 5 %. The absolute differences in flow values were small. Smaller differences were detected in the rest study than in stress studies.
Discussion and Conclusions: Previously, minimal differences between Q.Clear and VUE Point FX-S with a different PET/CT system and radiotracer in rest-only studies have been reported. In conclusion, Q.Clear and VUE Point FX-S on the Discovery MI PET/CT produce comparable MBF values in rest and stress studies of MBF with O-15 H2O with default beta value of 350.

Keywords: O-15 PET, reconstruction, Bayesian-penalized likelihood reconstruction
Poster panel: 232

Poster Number:

Reconstruction approach for partially truncated CT data (#1785)

K. Moothandassery Ramdevan1, D. Marlevi1, M. O'Regan2, M. Colarieti Tosti1

1 KTH Royal Institute of Technology, MTH/Medical Imaging, Stockholm, Sweden
2 Stockholm University, Faculty of Science, Department of Geological Sciences, Stockholm, Sweden


For various reasons it might be required to scan an object that partially lies outside the field of view(FOV) of a CT scanner. The parts of the object that lie outside the FOV will not contribute to the line integrals measured by the detector which causes image artifacts that affect the final image quality. In this paper, we suggest a novel reconstruction approach that estimates the attenuation by the object outside the FOV using a priori knowledge about the outline of the object. It is shown that, knowing the object's outline, it is possible to determine whether the attenuation along a given line is truncated. The total attenuation for a truncated projection is then estimated by interpolating the data between the consistent projections. The method therefore requires some of the projections to be consistent. This estimate, along with the knowledge of the distance traversed by the X-Ray inside the object is then used to determine the average attenuation. The method was tested on both numerical and physical phantoms and the results were encouraging. The performance was however only tested on simple and symmetrical objects such as Shepp-Logan phantom and objects with semi-circular cross section.

Keywords: computed tomography, projection truncation, truncation estimate
Poster panel: 235

Poster Number:

Resolution Modelling in Projection Space using Factorized Multi-block Detector Response Function (#1922)

H. Xu1, J. Scheins1, L. Caldeira1, M. Lenz1, B. Ma1, C. W. Lerche1, N. J. Shah1, 2

1 Forschungszentrum Juelich GmbH, Juelich, North Rhine-Westphalia, Germany
2 RWTH Aachen University, Aachen, North Rhine-Westphalia, Germany


Positron emission tomography (PET) images usually suffer from low spatial resolution and signal-to-noise (SNR) ratio. The degradation of image resolution in PET is caused by the complex detection process, e.g. inter-crystal scattering, crystal penetration. An accurate Detector Response Function (DRF) allows to model these phenomena and increases the spatial resolution as well as SNR in the iterative image reconstruction. However, fully 3D DRF for pixelated crystal arrays (block) which also considers inter-block penetration and inter-crystal scattering between different blocks still remains challenging. Here, we demonstrate the development of an accurate DRF for the Siemens Hybrid MR-BrainPET system with a 9-block model using GATE simulations. Different incident γ rays are described by four parameters (x, y, θ, φ) in Block Coordinate System. Their detection response, comprising a list of detected crystal IDs and corresponding detection probability, are stored as an entry of a 4D Look-up Table (LUT) addressed by (x, y, θ, φ). Based on the DRF LUT, a PSF blurring kernel in 4D projection space can be obtained by combining two multi-block DRF according to the intersected block pair for each Line-of-Response.

PSF modelling in projection space is implemented in the reconstruction toolkit PRESTO based on the developed DRF LUT. A resolution phantom with 6 types of hot rods is simulated by GATE and reconstructed by PRESTO with MLEM and MLEM-PSF. Visual results demonstrate that with moderate statistics (2.8×108), MLEM-PSF could recover small bins (5mm) at the edge of FOV in a more accurate way compared to MLEM. Furthermore, the images of MLEM-PSF show better noise suppresion.

Keywords: DRF, multi-block, resolution modelling, PRESTO, BrainPET
Poster panel: 238

Poster Number:

RISE: A method for 3D Reconstruction of SPECT Images - Validation with a Hardware Phantom (#1622)

L. Koutsantonis1, C. N. Papanicolas1, E. Stiliaris2, 1

1 The Cyprus Institute, CASTORC, Nicosia, Cyprus
2 University of Athens, Physics Department, Athens, Greece


The "Reconstructed Image from Simulations Ensemble" (RISE) has been recently demonstrated as an alternative method for tomographic image (2D) reconstruction in emission tomography, comparing favorably to the well-established techniques such as the MLEM and ART. Utilizing Monte-Carlo techniques, the method simulates an ensemble of image-solutions, each one corresponding to a different realization of the activity distribution. In this framework, the physical characteristics of the activity distribution are parametrized in a mathematical model which is employed to represent the imaged "object". The final reconstruction is derived by weighting the solutions satisfying a statistical criterion employed to quantify the "linkage" of the simulated image-solution to the set of projection data. In this work, we present the extension of RISE to a 3D reconstruction method directly providing the volumetric image of the activity distribution. The application of the 3D method in emission tomography is demonstrated with projection data obtained from a hardware 99m Tc phantom with a small γ -camera system. In this experiment, the method was examined in its capability to provide 3D reconstructions from a limited number of planar images (24) characterized by strong medium absorption.

Poster panel: 241

Poster Number:

A novel robust Cerenkov luminescence tomography method based on adaptive edge-preserving smoothing regularization strategy (#2137)

M. Cai1, Z. Zhang1, 2, Z. Hu1, J. Tian1

1 Chinese Academy of Sciences, Institute of Automation, Key Laboratory of Molecular Imaging, Beijing, China
2 Xidian University, School of Life Science and Technology, Xi’an, China


Cerenkov luminescence tomography (CLT) is a novel and potential approach for 3D imaging of radiopharmaceutical distributions for tumor detection and pharmacodynamics evaluation. However, it is a tremendous challenge to obtain accurate and robust reconstruction results because CLT is severely ill-posed and ill-conditioned resulting from the severe scattering effect of Cerenkov emission in biological tissues. It is well known that regularization strategies can be employed to improve the reconstruction accuracy of CLT. In this study, a novel modified adaptive edge-preserving smoothing regularization (AEPSR) was proposed and utilized to depict the edge of optical sources and reduce the noise in reconstruction of CLT. The AEPSR strategy contained four main parts: (1) self-example sub-dictionaries for optical sources were learned from the measured signals with the K-means clustering and principal component analysis (PCA) methods; (2) edge-preserving smoothing regularization term was used to preserve the edge of the reconstructed optical sources; (3) the smoothing regularization factor was adaptively determined; (4) the iterative shrinkage algorithm was used to solve the inverse problem. In vivo experiments were conducted in this study to validate the efficiency of AEPSR strategy for accurate and robust reconstruction of CLT. 4T1 subcutaneous tumor models (n=4) were constructed with female nude mice and the Cerenkov luminescence images were acquired after intravenous injection of 18F-FDG. By using of the AEPSR strategy, the reconstruction results of CLT can provide precise location of the tumors and the tumor shapes were able to be explicitly delineated. In conclusion, the AEPSR strategy reported here performed well in CLT and may contribute to the application of CLT in clinic.

Keywords: Cerenkov luminescence tomography (CLT), adaptive edge-preserving smoothing regularization (AEPSR), tumor detection, subcutaneous tumor models
Poster panel: 244

Poster Number:

A Self-supervised Deep Learning Network for Low-Dose CT Reconstruction (#2164)

K. Liang2, 2, L. Zhang2, Y. Yang1, H. Yang2, Y. Xing2

1 Tsinghua University, Engineering Physics, Beijing, China
2 Key Laboratory of Particle & Radiation Imaging, Engineering Physics, Beijing, China


Abstract-Low dose CT is of great interest in these days. Dose reduction raises noise level in projections and decrease image quality in reconstructions. Model-based iterative reconstructions can combine statistical noise model together with prior knowledges in an optimization problem so that significantly reduce noise and artefacts. In this work, we propose a deep learning neural network for low lose CT reconstruction by integrating the cost definition of iterative method so that a self-supervised learning can be done with no ground-truth information needed. Instead of minimizing cost function for each image, the network learns to minimize an ensemble cost function for the whole training set. Features learned can be robust. In this work, we implemented such a self-supervised deep learning network in the sense of penalized weighted least-squares (PWLS) together with a non-local mean constraint. We carried on our experiments on practical dental CT data. Plausible reconstructions with great noise reductions are obtained for test dataset non-overlap with training data. The image quality of test reconstructions are comparable to the results from the corresponding iterative reconstruction method. No iterative processes are needed for real data reconstruction with such a trained network. This method can be easily extended to other imaging systems and modalities.

Keywords: low-dose, PWLS, deep learing, reconstruction
Poster panel: 247

Poster Number:

A Dedicated Tomographic Image Reconstruction Algorithm for Integration-Mode Detector Configuration in Ion Imaging (#2257)

C. Seller Oria1, S. Meyer1, E. De Bernardi2, K. Parodi1, C. Gianoli1

1 Ludwig-Maximilians-Universität München, Department of Medical Physics, Experimental Physics, Garching, Bavaria, Germany
2 Università degli Studi di Milano-Bicocca, School of Medicine and Surgery, Milano, Italy


A dedicated tomographic image reconstruction algorithm for integration-mode detector configuration is proposed and investigated. The algorithm makes use of the stopping power (detector level) and the model of the scattering power (image level) for the entire histogram of the Water Equivalent Thickness (WET) components. Therefore, compared to conventional tomographic image reconstruction of the WET component with the maximum occurrence along straight trajectories, this algorithm is expected to provide better image quality.

An anthropomorphic phantom, adopted as ground truth, is used to simulate the proton trajectories. The system matrix of the proposed algorithm is based on the statistical estimation of the proton trajectories for each WET as a 2D Gaussian distribution according to a Multiple Coulomb Scattering model. The 2D Gaussian distributions are normalized to the unit, thus forming a 3D distribution, namely the “cone”, for each WET. The cones for each WET are interpolated to each integration line, thus adapting (stretching or dilating) the detector to the image. The updating formula of the Simultaneous Algebraic Reconstruction Technique is modified to account for the entire histogram of the WET components.

For computational reasons, the tomographic image reconstruction is implemented relying on ordered subsets of projection angles. With noise, tomographic image reconstruction benefits from an implementation with ordered subsets. In general, the proposed algorithm reports comparable median but smaller interquartile range in Root Mean Square Error of rWEPL than conventional tomographic image reconstruction.

The results indicate the potential of the proposed algorithm to provide better image quality than conventional tomographic image reconstruction. Future studies will include the application of the proposed algorithm to Monte Carlo simulations of clinical ion imaging based on both integration-mode and list-mode detector configurations.

Keywords: Ion Imaging, integration-mode detector configuration, stopping power, scattering power, Simultaneous Algebraic Reconstruction Technique
Poster panel: 250

Poster Number:

Joint reconstruction and flat-field estimation using support estimation (#2324)

N. Six1, J. De Beenhouwer1, V. Van Nieuwenhove1, W. Vanroose2, J. Sijbers1

1 University of Antwerp, Vision Lab, Department of Physics, Antwerp, Belgium
2 University of Antwerp, Department of Mathematics - Computer Science, Antwerp, Belgium


Reconstructions from X-ray computed tomography scans often suffer from ring artefacts. These artefacts are caused by non-uniformity in the response of different detector pixels. To reduce these artefacts, flat-fields are obtained and used to normalize each subsequent scan. Any discrepancies between the acquired flat-field and the true flat-field at the time of scanning will still lead to ring artefacts. We propose a novel method in the class of joint reconstruction and estimation algorithms, in which we simultaneously calculate a reconstruction and estimate a flat-field. The proposed method models the reconstruction-estimation problem as a single system of linear equations, which can be solved using fast Krylov methods. We used the Conjugate Gradients method (CG) for all reconstructions. We validated our proposed method by reconstructing a simulated dataset with a randomly generated non-uniform detector response. For the simulated data we calculated the error between the estimated flat-field and the ground truth flat-field to have a median of 0.0011% of the mean intensity. Next, we reconstructed an experimental dataset, where again a strong reduction of the ring artefacts was visible.

Poster panel: 253

Poster Number:

Dual-modality joint reconstruction of PET-MRI incorporating a cross-guided prior (#2384)

H. Li1, L. Lu1, S. Cao1, J. Gong1, Q. Feng1, A. Rahmim2, W. Chen1

1 Southern Medical University, School of Biomedical Engineering, Guangzhou, China
2 Johns Hopkins University, Baltimore, Maryland, United States of America

This work was supported by the National Natural Science Foundation of China under grants 61628105, 81501541, the National key research and development program under grant 2016YFC0104003, the Natural Science Foundation of Guangdong Province under grants 2016A030313577, and the Program of Pearl River Young Talents of Science and Technology in Guangzhou under grant 201610010011.


Combining PET and MRI, the hybrid PET/MR scanner can simultaneously acquire data from both the dual modalities. Therefore, the research on improving the quality of dual-modality images is necessary and important, and joint reconstruction of PET-MRI is worth studying so as to achieve superior images of both dual modalities synchronously. We therefore propose a cross-guided (CG) prior model based on guided functions, which can be incorporated into dual-modality joint reconstruction of PET and MRI by Maximum a posterior (MAP) framework. To test the feasibility and validity of the proposed CG prior, we compared it with separate reconstruction without prior (no prior), separate total variation (TV) prior, joint total variation (JTV) prior and linear parallel level sets (LPLS) prior. The objective functions based on different prior were solved with the same quasi-Newton method. The performance of different methods were verified with simulation data of 18F-FDG PET and T1-weighted MRI from brain phantom. As the results indicated, the proposed algorithm substantially reduced the noise and reconstruction error of both PET and MR images and achieved sharper and higher resolution images, which outperformed other state-of-the-art methods. In conclusion, the proposed CG prior is considerably effective, which can provide a potential way for dual-modality joint reconstruction.

Keywords: Joint Reconstruction, PET-MRI, cross-guided prior, dual-modality
Poster panel: 256

Poster Number:

A Preliminary Study on Optimization-Based Image Reconstruction from Sparse, List-Mode TOF-PET Data (#2451)

Z. Zhang1, S. Rose1, A. E. Perkins2, C. - M. Kao1, E. Y. Sidky1, X. Pan1, 3

1 The University of Chicago, Radiology, Chicago, Illinois, United States of America
2 Philips Healthcare, Cleveland, Ohio, United States of America
3 The University of Chicago, Radiation and Cellular Oncology, Chicago, Ohio, United States of America


There exists interest in designing a TOF-PET system with reduced detectors due to cost concerns, while not significantly compromising the utility. Previous work has demonstrated that advanced optimization-based algorithms have the potential for enabling the design of sparse-configuration non-TOF PET systems. In this study, we design and investigate a variety of TOF-PET configurations with 50% reduced number of detectors, and carry out preliminary studies with list-mode TOF-PET patient data corresponding to such sparse TOF-PET configurations. We in this work consider an optimization problem combining Kullback-Leibler (KL) data fidelity with an image TV constraint, and solve it by using a primal-dual optimization algorithm developed by Chambolle and Pock. Results show that appropriately designed algorithms may enable the design of innovative TOF-PET configurations with reduced number of detectors, while yielding potential practical clinical utilities.

Keywords: TOF PET, list mode, total variation, optimization-based reconstruction, sparse PET configuration
Poster panel: 259

Poster Number:

Investigation of Optimization-based Reconstruction from Sparse-view Data with Angularly-varying Truncation (#2556)

D. Xia1, Y. - B. Chang2, Z. Zhang1, B. Cheng1, J. Manak2, E. Y. Sidky1, X. Pan1

1 The University of Chicago, Radiology, Chicago, Illinois, United States of America
2 Canon Medical Research Institute USA, Vernon Hills, Illinois, United States of America


In neurological interventional and surgical procedures, measured data are generally truncated due to the limited detector size and the presence of some interventional devices outside of the imaging FOV of the C-arm CBCT, leading to prominent streak artifacts in clinical images especially when the scanning angular sampling is sparse. In this work, we investigate optimization-based image reconstruction from sparse-view, angularly-varying truncated data with reduced artifacts. The reconstruction problem is formulated as a constrained optimization program in which a combination fidelity term of data and data-derivative  was used for effective suppression of the truncation artifacts, and the Chambolle-Pock (CP) algorithm was used to solve the optimization program. The results of the study show that while the streak artifacts caused by the angularly-varying truncation in FDK reconstruction become severe as the number of view decreases, the image quality yielded by the optimization-based reconstruction remains largely unchanged.

Keywords: C-arm CBCT, Sparse-view, Optimization-based reconstruction, Angularly-varying truncation
Poster panel: 262

Poster Number:

Improved Dynamic PET Reconstruction Using Temporal Decoding (#2726)

S. Li2, 1, S. D. Jamadar1, 3, P. G. D. Ward3, 1, F. Sforazzini1, M. Premaratne2, G. F. Egan1, 3, Z. Chen1, 2

1 Monash University, Monash Biomedical Imaging, Clayton, Australia
2 Monash University, Department of Electrical and Computer System Engineering, Clayton, Australia
3 Monash University, Institute of Cognitive and Clinical Neuroscience, Clayton, Australia


Dynamic positron emission tomography (PET) plays a key role in recent continuous infusion FDG-PET (fPET) studies. However, the insufficient photon counts in short-time binning frame is a major obstacle in dynamic PET research. The aim of this paper is to design a dynamic PET reconstruction method by using the temporal information in projection domain. The hypothesis is that the PET reconstruction is an ill-posed problem. It is more efficient to reduce the spatial noise in projection domain than in the image space. This proposed method is achieved by two steps: (1) Dynamic PET reconstruction on a group of overlapped binning frames using a long window in projection domain. (2) A robust estimation method is used to estimate the reconstruction images corresponding to the short binning window from the data obtained in step 1. The purpose of the first step is to reduce the spatial noise of PET images, and step 2 corrects the temporal distortion introduced in in step 1.  A 2D dynamic reconstruction using the proposed method based on simulated constant infusion fPET human brain study is presented to demonstrate the effectiveness of the proposed method. The improvement of image qualities compared to the conventional independent frame reconstruction is clearly presented. The independent component analysis (ICA) can successfully detect the simulated activation in occipital regions using proposed method. On the contraryICA method failed to extract the meaningful information from conventional method. The proposed method shows a noticeable improvement in simulation study. The results could be further improved by using a more complex binning encoding method rather than the normal averaging window. This will be investigated in future work.


Keywords: Dynamic PET, Constant infusion
Poster panel: 265

Poster Number:

Improve 3D Cone-beam CT reconstruction by slice-wise deep learning (#2758)

H. Yang1, 2, K. Liang1, 2, L. Zhang1, 2, K. Kang1, 2, Y. Xing1, 2

1 Tsinghua University, Department of Engineering Physics, Beijing, China
2 Tsinghua University, Key Laboratory of Particle & Radiation Imaging, Ministry of Education, Beijing, China


As Computed Tomography (CT) technique develops fast, X-ray cone beam CT (CBCT) becomes widely used in many fields such as clinic diagnosis, public safety inspection and non-destructive testing. To reduce the potential radiation risk, lowering X-ray radiation dose becomes major concerns in these days. As computer hard-ware evolves greatly, researchers have devoted to optimization-based iterative methods when data is noisy. In the past few years, total variation (TV) regularization and nonlocal means-based (NLM) regularization have been used in image denoising. However, optimization-based iterative reconstruction is normally computational expensive. In these years, deep learning and convolutional neural network (CNN) is widely used in image processing areas. In the CT field, researchers have published some studies using CNN to reduce image noise for low-dose CT. In this work, we proposed a deep learning method for low-dose cone-beam CT reconstruction. By transferring 3D projection to a 2D problem with noise reduction property, we can not only obtain reconstruction of high image quality, but also lower the computation complexity from 3D projections to 2D. Our deep-learning network can significantly reduce the noise and improve the image quality. With the transformation of 3D cone-beam projection to 2D fan-beam projection, the reconstruction is greatly simplified. We will do more evaluation with real low-dose data and report our results in the conference.

Keywords: CT, CNN, Denoising, 3D projection, Low dose
Poster panel: 268

Poster Number:

Evaluation of a Block Matching and 3-D Collaborative Filter Algorithm for Dose Reduction in Molecular Breast Imaging (#1078)

M. K. O'Connor1, A. T. Tao1, A. L. Conners1, K. N. Hunt1, T. N. Swanson1, T. D. Tran1, A. D. A. Maidment4, L. R. Borges5, M. P. Johnson3, D. S. Lake2, A. Manduca2, C. B. Hruska1

1 Mayo Clinic, Radiology, Rochester, Minnesota, United States of America
2 Mayo Clinic, Physiology, Rochester, Minnesota, United States of America
3 Mayo Clinic, Biomedical Statistics, Rochester, Minnesota, United States of America
4 Univ. Pennsylvania, Radiology, Philadelphia, Pennsylvania, United States of America
5 Univ. Sao Paulo, Physics, San Paulo, Brazil


Background: Molecular breast imaging (MBI) employs dedicated dual-head gamma cameras for imaging uptake of Tc-99m sestamibi in the breast. MBI is performed with 300 MBq Tc-99m sestamibi (effective dose 2.1 mSv). Our goal was to validate the use of a BM3D algorithm to achieve 50% reduction in administered dose without compromising image quality.
Methods: MBI exams from 50 patients showing a focal mass (median size 12 mm; range 5-40 mm) with clinically-relevant count density (median 39 cts/pixel; range 29 - 63 cts/pixel) were selected. Images were acquired in dynamic mode (10 1-min frames). For each exam, three image types were generated: 1) standard image (sum of 10 frames), 2) “half-dose” (HD) image (sum of 5 frames), and 3) BM3D filtered half-dose (BM3D) image. A two-alternative forced choice task was performed independently by two breast imaging radiologists. For each patient, readers were presented with the standard image and either the HD or BM3D image in random order, while blinded to image type, and asked to select the image that best demonstrated the lesion. Lesion conspicuity on the selected image was assessed as a) the same, b) slightly better, or c) better than the non-selected image.
Results: The readers preferred standard to HD images (selected in 49/50 and 47/50 patients; both p<0.001), but showed no difference in preference between standard and BM3D images (selected in 27/50 and 23/50 patients, both p=0.67), Of the 100 comparisons (50 by each reader) between standard and BM3D images, lesion conspicuity was assessed as equivalent in 73% of images, slightly better in 12% of standard images and 12% of BM3D images, and better in 2% of standard images and 1% of BM3D images.
Conclusions: The BM3D algorithm was shown to be robust for a range of lesion sizes and count densities. Results suggest that with application of the BM3D algorithm, MBI can be performed with 150 MBq of Tc-99m sestamibi, reducing its effective dose to 1 mSv.

Keywords: molecular breast imaging, BM3D
Poster panel: 271

Poster Number:

Restoration of chest images by wavelet transform and convolutional neural networks (CNN) (#1187)

D. Lee1, S. Choi2, H. - J. Kim1, 2

1 Yonsei University, Department of Radiation Convergence Engineering, Research Institute of Health Science, Wonju, Republic of Korea
2 Yonsei University, Deparment of Radiological Science, Research Institue of Health Science, Wonju, Republic of Korea


High resolution images are necessary in many applications including medical imaging, nondestructive inspection, astronomy etc. Especially in medical imaging, images with high resolution are essential to detect precise anatomical structures. The most direct solution to increase spatial resolution in medical imaging is to use image sensor contained small pixel size and low signal modulation capability. However, as the technology of image sensor increases, the technical complexity and price increases rapidly, thus a signal processing technique that can improve the spatial resolution is needed. In this study, we developed a super resolution image processing algorithm based on deep learning and evaluated its performance. Proposed algorithm was developed by convergence of residual U-net based on convolutional neural network and wavelet transform. As the training data, we used the clinical data of chest radiograms provided by the cancer imaging archive (TCIA) collection. The proposed algorithm effectively restored the resolution degraded chest radiogram with bilinear interpolation. The structure similarity index measure between restored images and ground truth-images are over 0.95. The conspicuity of fine anatomical structures in chest such as terminal bronchiole was significantly improved by using proposed super resolution algorithm. In conclusion, we developed image processing algorithm that can reconstruct super resolution images from smoothed images. From the quantitative performance evaluation, the effect of the proposed algorithm was certained. This study is expected to contribute to the improvement of diagnostic efficiency through the implementation of super resolution images and to help accurate lesion targeting, segmentation and contouring.

Keywords: Deep learning, High resolution
Poster panel: 274

Poster Number:

Initial Investigation of Using a Generative Adversarial Network for Denoising in Dual Gating Myocardial Perfusion SPECT (#1349)

G. S. P. Mok1, 2, Q. Zhang1, X. Cun3, D. Zhang1, 2, H. Pretorius2, M. A. King2

1 University of Macau, Biomedical Imaging Laboratory (BIG), Department of Electrical and Computer Engineering, Faculty of Science and Technology, Macau, China
2 University of Massachusetts Medical School, Department of Radiology, Worcester, Massachusetts, United States of America
3 University of Macau, Department of Computer and Information Science, Macau, China


Dual respiratory-cardiac gating (DG) freezes myocardial perfusion SPECT by reducing both cardiac and respiratory motion. However, detected counts are divided into more bins in DG as compared to sole cardiac or respiratory gating, leading to further noise amplification. This study aims to reduce the noise level in each DG bin using a generative adversarial network (GAN). We used the XCAT phantom to generate a Tc-99m-MIBI DG dataset with cardiac and respiratory motion. A 5-s respiratory and a 1-s cardiac cycle were divided into 24 and 48 frames respectively, and then grouped to 6 respiratory and 8 cardiac gates, i.e., a total of 48 gates. We used an analytical projector to simulate a LEHR collimator with 120 realistic noisy projections over 180˚, which were reconstructed by OS-EM using average CT for attenuation correction. Eighteen DG bins, i.e., a total of 2052 images (18×114 axial slices), were paired with the corresponding single cardiac gate for training. Six other DGs of the same cardiac gate were tested by the GAN, which consisted of a 16-layer generator and a 5-layer discriminator implemented on a NVIDIA GTX 1080 GPU. The DG images w/ and w/o denoising were then registered to the end-expiration gate and summed to a cardiac image. The noise level, measured as the normalized standard deviation (NSD) on a 2D uniform region on the liver, was compared for a single cardiac gate (C8) and the registered cardiac image w/o (C8-REG) and w/ (C8-GAN) denoising. The FHWM measured across the left ventricle profile on the same images was used to assess the motion blurring. The NSD was 0.219, 0.229 and 0.093 and the FWHM was 2.51, 1.27 and 1.61 cm for C8, C8-REG and C8-GAN respectively. The noise level substantially reduced for the DG images when GAN was applied, with minimal increase of motion blurring. The GAN has potential to improve the DG performance thus clinical feasibility, and further evaluation on patient data is warranted to validate its actual clinical performance.

Keywords: SPECT, Dual-gating, deep learning, generative adversarial network
Poster panel: 277

Poster Number:

High resolution computed tomographic imaging by wavelet transform and convolutional neural networks (CNN) (#1504)

D. Lee1, S. Choi2, H. - J. Kim1, 2

1 Yonsei University, Department of Radiation Convergence Engineering, Institute of Health Science, Wonju, Republic of Korea
2 Yonsei University, Department of Radiological Sciecne, Institute of Health Science, Wonju, Republic of Korea


Computed tomographic images with high resolution are necessary for accurate diagnosis. To improve image resolution in computed tomographic imaging, small pixel/voxel size in space and low signal modulation capability of detector are the most direct solution. In this case, the price of detector increases in proportion to the detector performance, and the smaller the pixel/voxel size resulted in the longer the reconstruction time. It is alternative to improve the image resolution by machine learning technique. We used deep learning based technology to improve image resolution of computed tomographic images. In particular, proposed deep learning algorithm was developed by convergence of residual U-net based on convolutional neural network and wavelet transform. As the training data, we used the clinical data of computed tomographic images provided by the cancer imaging archive (TCIA) collection. Developed algorithm based on deep learning successfully restored high resolution components. By summating predicted images from low resolution computed tomographic images, high resolution computed tomographic images were acquired. The structure similarity index measure (SSIM) of predicted images were improved compared to SSIM value of the low resolution computed tomographic images. In conclusion, we developed a algorithm for improving image resolution in computed tomographic imaging. The results are not only expected to contribute enhancing image quality, but also expected to reduce reconstruction time of diagnostic computed tomographic images.

Keywords: High resolution, Deep learning
Poster panel: 280

Poster Number:

Automatic MismatcAutomatic Mismatch Correction and Motion compensation for Free-breathing PET/CTh Correction and Motion compensation for Free-breathing PET/CT (#1616)

J. Wang1, T. Feng1, Y. Dong1, H. Li1

1 UIH America, Inc., Houston, Texas, United States of America


In clinical free-breathing(FB) PET/CT imaging protocol, a single CT image of the patient is obtained before PET scan, which is inadequate for attenuation correction of the FB PET data due to respiratory motion. The aim of this study is to design an automatic workflow to correct mismatch and reduce motion blurring of FB PET and single-scan CT data.

First, the PET data was divided into four equal-count respiratory frames using data-driven respiratory gating method. Modified attenuation map was used in during gated reconstruction to obtain mismatch-free gated PET images for motion estimation purpose. Second, the respiratory phase of the original CT image was identified automatically using mutual information (MI) as similarity metric. Two regions of interests (ROIs) near the lungs were segmented from CT image. The MI inside the ROIs of each gated PET reconstruction and the CT image was measured for all gates and the phase with highest MI value is identified as the correct respiratory phase of CT image. Third, using the respiratory phase of the CT image as reference gate, motion vector fields (MVFs) from every other gate to the reference gate were estimated using a B-spline based multi-resolution image registration algorithm. Fourth, the original CT image was transformed to every other frame using estimated MVFs to obtain phase-matched attenuation maps. Finally, image reconstruction of gated PET data was repeated using phase-matched gated PET-CT data pairs. MVFs estimated in step 3 were applied to this set of gated PET images to obtain final motion compensated PET image.

This method was applied to clinical datasets of F18-FDG and Ga68-DOTA-NOC tracer. The proposed method successfully identified respiratory phase of the CT image and generated phase-matched CT images for all respiratory phases of PET data. Final reconstructed PET data using the proposed method showed much reduced mismatch artifacts and motion blurring, as well as improved statistics.

Keywords: mismatch correction, motion compensation
Poster panel: 283

Poster Number:

Improving time-of-flight performance of BGO-SiPM detectors with convolutional neural networks (#1690)

E. Berg1, S. I. Kwon1, S. R. Cherry1

1 University of California, Davis, Biomedical Engineering, Davis, California, United States of America


Bismuth germanate (BGO) has gained interest for time-of-flight (TOF) PET by exploiting its relatively high Cerenkov photon yield. Excellent FWHM coincidence timing resolution has recently been achieved (<400 ps), however the influence of both Cerenkov and scintillation light on the timing estimate leads to large FWTM values (~2.5 ns). We previously developed a method for estimating TOF directly from coincidence pairs of digitized waveforms using convolutional neural networks (CNNs), and demonstrated a ~20% improvement in timing resolution compared to leading edge discrimination when using 10 mm LYSO crystals coupled to PMTs. Here we demonstrate CNN-based timing with BGO crystals coupled to SiPMs. Two identical detectors were used, each consisting of a 3 x 3 x 20 mm Teflon wrapped BGO crystal coupled to a 4 x 4 mm NUV-HD SiPM (FBK). SiPM signals are passed through an amplifier that provides two outputs: a “slow” signal, and a “fast” signal obtained by filtering with a pole-zero cancellation circuit, and digitized at 50 GS/s. We compared two CNN methods: using only the fast signals, or using both the fast and slow signals simultaneously in the CNN. Both methods used the same overall CNN architecture (3 convolution layers, 2 fully connected layers, and regression output). Leading edge discrimination on the fast waveforms yielded a coincidence timing resolution of 492 ps FWHM (1450 ps FWTM), in agreement with previous measurements with a similar detector, compared to 431 ps FWHM (1271 ps FWTM) using a CNN on the fast signals and 371 ps FWHM (1136 ps FWTM) using a CNN on both fast and slow signals simultaneously. All results were obtained with a 430 – 590 keV energy window. First, this study demonstrated a method to improve timing resolution of BGO-based detectors. Second, by extending CNN-based TOF estimation to (i) 20 mm long crystals, (ii) SiPMs, and (iii) BGO with its non-traditional prompt Cerenkov signal, we demonstrated the broader utility of CNNs for TOF PET.

Keywords: time-of-flight, neural network, bismuth germanate (BGO), positron emission tomography (PET), Cerenkov
Poster panel: 286

Poster Number:

Image Quality Determination of a Novel Low Energy X-ray Detector (#1796)

H. M. Alzahrani1, S. Richards2, I. Sedgwick2, P. Seller2, A. C. Konstantinidis3, G. Royle1, K. Ricketts4

1 University College London, Department of Medical Physics & Biomedical Engineering, London , United Kingdom
2 Science and Technology Facilities Council, Rutherford Appleton Laboratory, Didcot, United Kingdom
3 Christie Medical Physics and Engineering, The Christie NHS Foundation Trust, Diagnostic Radiology and Radiation Protection, Manchester, United Kingdom
4 University College London, Division of Surgery and Interventional Science, London , United Kingdom

This is a collaborative project between Rutherford Appleton Laboratory (Oxford, UK) and University College London (London, UK).


The demand for adequate image quality with low radiation doses for patients has greatly increased. This is especially true in the case of position verification in radiotherapy which requires a high number of images per patient. This study presents a physical characterisation of a new clinical detector based on a thick layer (1 mm) of structured Cesium Iodide and complimentary metal-oxide-semiconductor technology with active pixel sensor architecture. The detector size is 32.4 ×21.4 cm² with an active area of 24 x14 cm² and a pixel pitch of 50µm. It was developed by Rutherford Appleton Laboratory (Oxford, UK) for general x-ray and cone beam computed tomography (CBCT) applications. We made a critical appraisal of its performance for the first time and determined the dark current, signal transfer property (STP) and its detective quantum efficiency (DQE) by acquiring the pre-sampling modulation transfer function and normalised noise power spectrum. The investigation was conducted with the application of three x-ray beam qualities: (50 kV (RQA3), 70 kV (RQA5) and 90 kV (RQA7)) in compliance with the International Electrotechnical Commission (2003) standard. The dark current was 2.51 pA/cm²; in equivalent terms, the detector accumulated 391.3 e-/s. As a result, the integration time of 0.13 s was selected to diminish the contribution of dark noise signal in the signal-to-noise ratio (SNR) measurements. The STP was found to be reasonably linear with the coefficient of determination (R²) greater than 0.9995. A relatively low degree of noise was evident. It also has a limiting resolution of 3 line pairs per millimetre (lp/mm). The DQE values at 0.5 lp/mm, representing the best-case scenario, were 0.46 for RQA3, 0.52-0.56 for RQA5 and 0.55-0.59 for RQA7 under low exposures (0.26-2µGy). Increasing the energy increased the DQE due to SNR enhancement. Hence, the most energetic beam (90kV), which is used for CBCT, presented higher resolution and lower noise leading to higher DQE.

Keywords: signal-to-noise ratio, image quality, DQE, CBCT, detector characterisation
Poster panel: 289

Poster Number:

High Accuracy Automated Diagnosis of Parkinson's Disease (#1859)

I. Ozsahin1, B. Sekeroglu2, G. Altinoglu1

1 Near East University, Department of Biomedical Engineering, Nicosia, Turkey
2 Near East University, Department of Information Systems Engineering, Nicosia, Turkey


Parkinson's disease, which is the second most common neurodegenerative disease, can be diagnosed clinically when about 70% of dopaminergic neurons are lost and symptoms are noticed. Neuroimaging methods such as single photon emission tomography have been a useful tool in vivo to assess dopamine transporters in striatal region. However, inter- and intra-reader variability of construing the images might result in misdiagnosis of the patients. To overcome the challenge posed by classification of the disease, artificial neural networks have been proposed. In this study, we used back propagation neural network on dopamine transporter imaging data sets from Parkinson’s Progression Markers Initiative. Uniquely, we found the sensitivity, specificity, and accuracy for Parkinson's disease vs. control group classification as 99.70%, 99.24%, 99.64%, respectively, the highest values achieved in the existing literature to our knowledge.

Keywords: Parkinson
Poster panel: 292

Poster Number:

A Novel Three-Dimensional Clustering Method Incorporated with Ellipsoid Model Fitting for Personalizing Myocardial Perfusion Positron Emission Tomography Image Segmentation (#1907)

J. - Y. Sun1, 2, T. - H. Wu1, 2, H. Liu2, Y. - H. Liu1, 2

1 National Yang-Ming University, Department of Biomedical Imaging and Radiological Sciences, Taipei, Taiwan
2 Yale University, Department of Internal Medicine (Cardiology), New Haven, Connecticut, United States of America


Abstract The quantification of myocardial blood flow (MBF) from cardiac positron emission tomography (PET) has been used for evaluating the abnormalities of coronary flow reserve for patients with known or suspected microvascular disease. We proposed a novel approach to automatic segmentation of the left ventricle (LV) and generating a personalized myocardial model from 82Rb myocardial perfusion positron emission tomography (PET) images using a scheme of 3-dimension (3-D) clustering and ellipsoid model fitting. The new methods were evaluated using 17 patients with normal myocardial perfusion (normal group) and 20 patients with myocardial perfusion defect (abnormal group) who underwent rest and stress 82Rb PET imaging. For comparisons, the PET images were processed using k-means with 4 clusters (k-ms4), fuzzy c-means with 4 clusters (FCM4), and fuzzy c-means with 3 clusters (FCM3). The myocardial mass’ of both rest and stress studies of each patient was calculated. Paired t-test was used for the assessment of statistical significance in the difference between two measures and the Bland-Altman analysis was used to evaluate the agreement between the two measures. The personized myocardial model was successfully generated for all the 37 patients included in this study. The myocardial mass’ derived from FCM3 were higher than those estimated using the other methods being compared in this study. There were no significant differences between the k-ms4-derived and the FCM4-derived myocardial mass’, whereas there were significant differences between the k-ms4-derived and FCM3-derived myocardial mass’. The myocardial mass’ calculated from FCM4 and FCM3 were also significantly different. Thus, our methods may have a potential of automatically segmenting the LV myocardial regions from the myocardial perfusion PET images, whereas the performance of the new automated methods can be limited by the partial volume effect of the heart with a small LV cavity.

Keywords: PET, 82Rb, cardiac, segmentation, clustering
Poster panel: 295

Poster Number:

Automatic detection of the atlas profile in CT images of Amyothrophic Lateral Sclerosis (ALS) patients (#1929)

A. M. Massone1, 3, C. Campi2, M. C. Beltrametti3, M. Piana3, 1, C. Marini4, G. Sambuceti5

1 CNR - SPIN, Genova, Italy
2 Università degli Studi di Padova, Dipartimento di Medicina, Padova, Italy
3 Università degli Studi di Genova, Dipartimento di Matematica, Genova, Italy
4 CNR-IBFM, Milano, Italy
5 IRCCS Ospedale Policlinico San Martino, Medicina Nucleare, Genova, Italy


Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder characterized by a degeneration of upper and lower motor neurons. Recent PET/CT analysis indicated that the motor/premotor cortex has a prevalent glucose hypometabolism evaluated by 18F-2-D-fluorodeoxyglucose (FDG) uptake in PET/CT studies. Further, we have recently presented an imaging technique based on the use of the Hough Transform (HT) for the segmentation of the spinal canal and spinal cord (SC) in X-ray CT images. Information inferred from CT images have been integrated with functional information from PET data to evaluate the metabolic activity of the spinal marrow in control subjects and ALS patients. Our method allowed the evaluation of cervical SC hypermetabolism with respect to controls. A simultaneous analysis of brain and SC FDG uptake showed that cervical SC activity was inversely correlated with activity in the pre- and post-central gyri, particularly in the left hemisphere. Similarly, dorsal SC SUV was inversely correlated with activity in the temporolateral and frontal dorso-lateral cortex in the left hemisphere. The automatic detection of the SC profile in CT images was realized by first automatically recognizing the spinal canal profile, which allowed the identification of regions of interest where the SC profile can be more easily identified. Also the recognition of the spinal canal profile was performed by means of the HT method. In that analysis the cervical segment was defined as the region between skull basis and the plane adjacent to the caudal face of C7 vertebral body but slices corresponding to the C1 vertebra (atlas) were always excluded from this kind of analysis because they are difficult to manage with the HT tool. This paper shows how to obtain an effective segmentation of the atlas, which can lead to evaluate the contribution of FDG metabolism from the superior cervical spine district to the interplay between spinal cord and cortical metabolism in ALS.

Keywords: Amyotrophic lateral sclerosis, PET/CT data, Pattern recognition, Hough transform
Poster panel: 298

Poster Number:

Voxel-based partial volume correction of amyloid PET images incorporating non-local means regularization (#1981)

Y. Gao1, H. Zhang2, Y. Zhu3, M. Bilgel4, O. Rousset3, S. Resnick4, D. F. Wong3, L. Lu1, A. Rahmim3

1 Southern Medical University, Department of Biomedical Engineering, Guangzhou, China
2 Johns Hopkins University, Department of Biomedical Engineering, Baltimore, Maryland, United States of America
3 Johns Hopkins University, Department of Radiology, Baltimore, Maryland, United States of America
4 National Institute on Aging, Baltimore, Maryland, United States of America


Amyloid PET imaging is increasingly utilized to assess Alzheimer’s disease. Nonetheless, PET imaging can be significantly degraded by the partial volume effect (PVE). This issue has been tackled via a number of post-reconstruction partial volume correction (PVC) methods. In our work, we proposed a voxel-based PVC method using non-local means (NLM) regularization under the weighted least squares framework that models the point-spread function of the PET system. The NLM algorithm has been proposed to suppress image noise while preserving edge information for natural images. This algorithm utilizes the high degree of information redundancy that typically exists in images, and reduces image noise by replacing each pixel intensity with a weighted average of its non-local neighbors. Based on its advanced property, we propose to employ NLM as a regularization term in PET PVC. For a penalized weighted least squares (PWLS) objective function, we used the Gauss-Seidel (GS) optimization algorithm, regularized using a one-step-late (OSL) framework. Under assumption of independent, identically-distributed (iid) Gaussian noise, then the PWLS framework becomes standard least squares. When the steepest descent scheme is applied to the problem, this leads to the iterative ‘reblurred’ Van Citter (VC) method. We tried both the VC method, and GS which involves a more sophisticated step-size method. In any case, the iid assumption is especially violated in OSEM reconstruction where the variance image is roughly proportional to the image (thus not uniform as in FBP). In the present work, we assessed the impact of appropriate variance weighting, as well as added NLM regularization. Our results demonstrate that statistical weighting improved quantitative bias vs. noise performance; and also, NLM regularization method exhibits improved performance. These were especially the case in the small regions relevant in Alzheimer’s disease research.

Keywords: Amyloid PET imaging, partial volume correction, non-local means regularization, penalized weighted least squares
Poster panel: 301

Poster Number:

Scan Time Reduction of Neuro FDG PET using Deep Learning (#2116)

J. Kim1, S. Kang1, K. Lee1, J. H. Jung1, H. K. Lim1, G. Kim1, S. Lee2, M. Yun2, Y. Choi1

1 Sogang University, Department of Electronic Engineering, Seoul, Republic of Korea
2 Yonsei University College of Medicine, Departments of Nuclear Medicine, Severance Hospital, Seoul, Republic of Korea


PET requires long scan time and sufficient amount of radiotracer injection to acquire high quality image causing patient's inconvenience and radiation exposure which motivates to reduce scan time and/or injected dose. The purpose of this study was to predict FDG PET image (T-PET) of human brain having high signal-to-noise ratio (SNR) acquired for typical scan duration from short scan FDG PET image (S-PET) with low SNR using deep learning method to reduce the scan time. U-net based convolutional neural network (CNN) with concatenate connection between corresponding stage was implemented. Residual learning framework between input and output of the network and fully connected layer between encoder and decoder were added to improve the quality of predicted T-PET. The proposed network was trained using neuro FDG PET of 23 normal subjects and then tested using data of 4 normal subjects independent from the training data. T-PETs were generated using listmode data acquired for 900 s with intravenous injection of about 170 MBq of 18F-FDG. S-PETs were generated using listmode data acquired for 10, 30, 60 and 120 s to evaluate the accuracy of the proposed CNN as a function of scan time. Predicted T-PETs were successfully generated using S-PETs of 10 s although modest degradations of spatial resolution were observed in the particular brain region such as middle frontal and temporal gyrus. The qualities of predicted images were gradually improved with increasing scan time and no noticeable difference between ground-truth and predicted T-PETs was observed with S-PETs of 120 s. The normalized root mean square errors and peak signal to noise ratios between ground-truth T-PETs and predicted T-PETs obtained using S-PETs of 10, 30, 60, 120 s were 0.13/23.66, 0.10/25.93, 0.08/27.98 and 0.06/30.61, respectively. Experimental results indicate that it is possible to generate T-PET from neuro FDG S-PET using the proposed CNN method which allows to reduce scan time as well as injected dose.

Keywords: deep learning, FDG PET, scan time reduction
Poster panel: 304

Poster Number:

Development of deep learning model for prediction of chemotherapy response using PET images and radiomics features (#2166)

W. Kim1, J. Park1, H. Sheen1, B. H. Byun2, I. Lim2, C. - B. Kong3, S. M. Lim2, S. - K. Woo1

1 Korea Institute of Radiological & Medical Science (KIRAMS), Division of Radiation Biomedical Research, Seoul, Republic of Korea
2 Korea Institute of Radiological & Medical Science (KIRAMS), Department of Nuclear Medicine, Seoul, Republic of Korea
3 Korea Institute of Radiological & Medical Science (KIRAMS), Department of Orthopedic Surgery, Seoul, Republic of Korea


Convolution neural network (CNN) was consider the input images, and then it could be resulted in less accuracy or incorrect prediction of therapeutic response. In order to improve the weakness, we development of prediction model using advanced deep learning techniques of chemotherapy response. The osteosarcoma patients were separated by two groups, responder (n=28) and non-responder (n=26). 18F-FDG PET images were acquired in osteosarcoma patients (n=54) before the chemotherapy. The acquire data have been analyzed by radiomics. Selected radiomics features were using information gain (IG) method. The different types of deep learning prediction models were compared as following: 1) with radiomics features model that is advanced CNN (A-CNN), 2) without radiomics feature model that is CNN. A-CNN and CNN model has 2 convolution layer and max pooling layers. Convolution and max pooling layers were generated the feature map. A-CNN was concatenated the feature map value and radiomics feature before the fully connected layer. Total seventy radiomics features were extracted from 18F-FDG PET images of osteosarcoma patients before the chemotherapy. Nine radiomics features of which IG value was upper 0.1 were selected and suggested for CNN. The A-CNN 70, 3, 9 radiomics feature prediction accuracy was 81, 82, 85%, while CNN was 80%. The results of the chosen nine radiomics features provided the more increased 5% accurate prediction than CNN model. In this study, we development chemotherapy prediction model using advanced deep learning technique. Selected radiomics features and PET image deep learning model has been shown more accurate result than 70 radiomics feature of A-CNN and CNN models. This advanced deep learning model will help to consider the radiomics feature as well as disease related feature such as genomics data.

Keywords: Convolution neural network, Selected radiomics features, Chemotherapy prediction model, Osteosarcoma, PET
Poster panel: 307

Poster Number:

Contrast CT image generation model using CT image of PET/CT (#2208)

W. Kim1, S. - K. Woo1, J. Park1, H. Sheen1, I. Lim2, S. M. Lim2, B. H. Byun2

1 Korea Institute of Radiological & Medical Science (KIRAMS), Division of Radiation Biomedical Research, Seoul, Republic of Korea
2 Korea Institute of Radiological & Medical Science (KIRAMS), Department of Nuclear Medicine, Seoul, Republic of Korea


Contrast CT imaging is a useful method to highlight structures such as blood vessels and tissue. The aim of this study was to generate contrast enhanced CT image for defining region of blood vessels and tumor from PET/CT image using deep learning approach. The deep learning approach in this study was based on conditional generative adversarial network (cGAN). The structure of cGAN composed of two convolutional neural network (CNN), generator and discriminator. Generator was trained to generate contrast CT (G-CCT) image from CT image which cannot be discriminated from the original contrast CT (CCT) image. Discriminator was trained to discriminate CCT image from G-CCT image by generator. To improve the generation performance of the model, we applied adaptive histogram equalization as image preprocessing. We compare the model trained with CT and the model with histogram equalized CT. Structural similarity index measurement (SSIM) was calculated to measure the similarity between G-CCT and CCT image. We defined and compared the region of lymphoma in left upper cervical lymph node level 2 using CT, CCT, and G-CCT. Detect the lymph node cancer was calculated by short to long axis ratio (S/L) method. In the case of the model trained with CT, SSIM was 0.9493±0.0183 in training phase and 0.9055±0.0484 in testing phase. In the case of the model trained with histogram equalized CT, SSIM was 0.954±0.0149 in training phase and 0.9081±0.047 in testing phase. G-CCT images showed more enhanced contrast in blood vessels and tumor region than CT image. Lymph node  cancer S/L of CT, CCT, and G-CCT were 0.412, 0.395, and 0.397, respectively. The  tumor S/L of generated contrast CT image was evaluated similar to these of real contrast CT image. This contrast CT image generation based on deep learning might be able to suggest a cost-effective and less-hazardous process of acquiring contrast CT image to patients as well as more anatomical information with only CT scan.

Keywords: Contrast CT, Conditional generative adversarial network, Lymph node, CT, Adaptive histogram equalization
Poster panel: 310

Poster Number:

A deep features-based radiomics model for breast lesion classification on FFDM (#2401)

C. Liang1, 2, Z. Bian1, 2, W. Lyu1, 2, D. Zeng1, 2, J. Ma1, 2

1 Southern Medical University, School of Biomedical Engineering, Guangzhou, China
2 Southern Medical University, Guangzhou Key Laboratory of Medical Radiation Imaging and Detection Technology, Guangzhou, China

This work was supported in part by the National Natural Science Foundation of China under Grant U1708261, Grant 81701690, Grant 61701217 and Grant 61571214, the China Postdoctoral Science Foundation funded project under Grant 2016M602489, the Guangdong Natural Science Foundation under Grant 2015A030313271, the Science and Technology Program of Guangzhou, China under Grant 201510010039, and the Science and Technology Program of Guangdong, China under Grant 2015B020233008.


The radiomics model can be used in breast cancer detection via calculating texture features in the lesion. However, the texture features are explicitly designed or handcrafted in advance, and this would limit their ability to characterize the lesion properly. This paper aims to build a deep features-based radiomics model to classify benign and malignant breast lesions on full-field digital mammography (FFDM). Specifically, the presented model considers the texture features learned from the deep learning network. This study consists of 106 retrospective data in both craniocaudal (CC) view and mediolateral oblique (MLO) view. First, 23 handcrafted features (HCF) are extracted from breast lesion, and 4096 deep features (DF) are extracted from the pre-trained deep learning model. Given that CC view and MLO view provide different breast lesion information, we consider to combine two extracted features as combined-views. After T-test selection, a suitable feature set of hand-crafted features is selected. Finally, a multi-classifiers model is trained on the combination of HCF and DF. The experiment results demonstrate that the presented model can achieve better classification performance (AUC=0.946) compared with HCF only (AUC=0.902) and DF only (AUC=0.832).

Keywords: Radiomics, breast cancer detection, texture features, lesion classification, full-field digital mammography
Poster panel: 313

Poster Number:

Prediction of Outcome in Parkinson’s Disease Patients from DAT SPECT images using a convolutional neural network (#2560)

M. P. Adams1, B. Yang1, A. Rahmim2, J. Tang1

1 Oakland University, Department of Electrical and Computer Engineering, Rochester, Michigan, United States of America
2 Johns Hopkins University, Department of Radiology and Radiological Science, Baltimore, Maryland, United States of America


Dopamine transporter (DAT) SPECT imaging is widely used for the diagnosis of Parkinson’s disease (PD). Learning-based techniques have achieved promising results for the use of DAT SPECT images in improving outcome predictions. The purpose of this study is to use the convolutional neural network (CNN) in predicting the motor part (III) of the unified Parkinson’s disease rating scale (UPDRS) score group for a given patient. We cast the motor function score prediction as a categorization problem that predicts whether a given patient will have a score above or below 30 in 4 years from baseline. From the PPMI database, we gathered motor function score and DAT SPECT pairs of 293 subjects with PD, in which 90% are used for training and 10% for testing. The CNN was trained with the baseline image and score and the corresponding year 4 score category of the training subjects using a backpropagation algorithm. The remaining subjects were then used to test the trained CNN to predict the motor function categories. This procedure was repeated to perform 10-fold stratified cross validation to ensure that the CNN was tested on all the subjects. It was made sure that all 10 folds be representative of the 293 subject population’s proportion in each score category. To quantitatively evaluate the prediction results, the mean percent accuracy across the 10 rounds of 10-fold cross validation was calculated to be 67.1+/-8.0%. This procedure was then repeated following the same method while omitting all the image data, which resulted in an accuracy of 64.5+/-8.1%. The two-sample t-test on these accuracy results yielded a two-tailed p-value of 0.02, which shows statistically significant improvement when the imaging data was used. This CNN-based 4 year UPDRSIII score predicting scheme demonstrates the contribution of DAT SPECT images, which shows their promise of improving progression tracking for patients with PD.

Keywords: DAT SPECT, Deep Learning, Convolutional Neural Network, Parkinson's disease, Outcome prediction
Poster panel: 316

Poster Number:

Feature-Oriented Deep Convolutional Neural Network for PET Image Denoising (#2629)

C. Chan1, J. Zhou1, L. Yang1, W. Qi1, J. Kolthammer1, E. Asma1

1 Canon Medical Research USA, Vernon Hills, Illinois, United States of America


Deep convolutional neural networks have demonstrated superior performance in natural image denoising. Trainable network weights have been typically optimized by minimizing a loss function that computes pixel-wise discrepancies between the noisy image and the clean target image. In this study, we investigate an alternative solution that utilizes more prior knowledge to highlight the features of interest by modifying their contributions to the global loss function. We propose a feature-oriented deep convolutional neural network (FeaOri-DCNN) for PET image denoising that uses weight maps in order to steer the training toward contrast preservation for small features. To obtain the weight maps, we first manually segment the lesions in the target images to create lesion masks. Lesion voxels are then assigned weights equal to the ratio between the size of the background and the size of the lesion. This weight map is then incorporated into the loss function optimization. We trained the proposed FeaOri-DCNN and a conventional DCNN built on a five-layer residual network architecture with simulated and phantom images containing hot spheres with various size and contrast. The results show that FeaOri-DCNN improved contrast recovery on small and low contrast spheres by up to 36% while performing similarly in terms of noise reduction in the background.

Keywords: deep learning, whole-body PET, denoising, convolutional neural network
Poster panel: 319

Poster Number:

Investigation of transfer learning training dataset using realistic simulated lesion models in the application of convolutional neural network to classification of pancreatic lesions (#2742)

T. - S. Lee1, J. Xu1, B. M. W. Tsui1

1 Johns Hopkins University, Radiology, Baltimore, Maryland, United States of America


Transfer learning methodology has been used to provide pretrained convolutional neural network to improve accuracy especially in medical imaging applications where a large number of clinical images with known ‘truth’ is difficult and expensive to obtain. The aim of this work is to investigate the requirements and characteristics of training dataset for the transfer learning using realistic simulated lesion models in the classification of pancreatic lesions. We utilized Inception V3, a deep learning architecture from Google that had already been trained by a huge collection of generic images. We simulated images of five models of pancreatic cystic lesions with wide ranges of lesion contrasts, image noise levels, and number of images for each lesion model and contrast and noise variation. Ninety percent of the total number of simulated images were used in the training and ten percent were used in testing. When the training image set consisted of all the variations and the testing images are within the same range of the variations, the classification accuracy remained constant from a total number of 49,000 down to 13,000 images. When the training image set and the test images consisted of a range of lower contrast and higher noise level, the accuracy was found to be lower than that obtained with a range of higher contrast and higher noise level. When the test images are outside the range of the training image set, the classification accuracy was found to be lower than when they are inside the range. We conclude the classification accuracy of the transfer learning method is highly dependent on the characteristics of the training dataset in the classification of pancreatic lesions. An optimally designed training image dataset plays an important role in the effectiveness of the transfer learning method with pretrained deep learning model in lesion classification in medical imaging.

Keywords: convolutional neural network, transfer learning, classification, pancreatic lesion, simulation
Poster panel: 322

Poster Number:

3D probability driven random walk segmentation with automated seed selection for the delineation of PET volumes (#2914)

T. M. S. - M. Osman1, B. McBride2, T. Hennessy2, S. Downes2, A. B. Rozenfeld1, A. Malaroda1

1 University of Wollongong, Centre of Medical Radiation Physics, School of Physics, Wollongong, Australia
2 Prince of Wales Hospital, Randwick, Australia


Rapid, robust and reliable delineation of tumours is crucial for the radiation therapy treatment planning process, especially when PET volumes are used in conjunction with CT. The aim of this study was to investigate the performance, on both static and gated PET images, of a 3D random walk image segmentation algorithm, which makes use of an automated seed selection method and an adaptive probability threshold (3DAARW). This segmentation algorithm was compared to the performance of a published 2D adaptive probability threshold counterpart (2DAARW) as well as the original 2D and 3D random walk algorithms (2DARW and 3DARW) with fixed probability thresholds utilising the same automated seed selection methodology as the 2DAARW and 3DAARW. Static and gated PET/CT images of a moving NEMA IEC phantom were acquired, with six 18F-FDG filled spheres of diameters ranging from 10mm to 37mm and 3 source-to-background ratios (9:1, 6:1 and 3:1). In order to optimise the method over multiple source-to-background ratios and sphere sizes, the segmentation parameters were systematically varied searching for the optimal combination of parameters minimising the overall percentage difference. The performance of each method was then assessed by calculating the Dice Coefficient of the segmented PET volumes with the ground truth CT. In general, under the conditions investigated, 3D RW performed better than 2D RW and, in particular, the 3DAARW performed best on the static images, yielding a median DC of ~0.85, whilst the 3DARW performed best on the gated images, yielding a median DC of ~0.88 with 75% of the values above 0.85.

Keywords: Image segmentation, Random Walk, 4D PET/CT, Radiation therapy treatment planning
Poster panel: 325

Poster Number:

A New Method for Quantification of Left Ventricular Function from Low-dose Equilibrium Radionuclide Angiocardiography: Comparisons with Conventional Methods in Patients (#1085)

Y. - H. Liu1, 4, V. Tsatkin2, R. Fazzone-Chettiar2, E. J. Miller1, 3, A. J. Sinusas1, 3

1 Yale University, Internal Medicine (Cardiology), New Haven, Connecticut, United States of America
2 Yale New Haven Hospital, Nuclear Cardiology, New Haven, Connecticut, United States of America
3 Yale University, Radiology & Biomedical Imaging, New Haven, Connecticut, United States of America
4 Chung Yuan Christian University, Biomedical Engineering, Taoyuan, Taiwan


Planar equilibrium radionuclide angiocardiography (ERNA) has been used as a gold standard for assessments of the left ventricular (LV) function with high precision for over 3 decades. However, this imaging modality has recently gained less favor due to the growing concerns of radiation exposure to patients. Although single photon emission computerized tomography (SPECT) cameras with high-sensitivity detectors and high-resolution collimators can acquire images with much-reduced radiation dose and/or acquisition time, calculations of LV end-diastolic volume (EDV), ejection fraction (EF) and peak filling rate (PFR) from gated blood pool SPECT require a reliable LV segmentation. Furthermore, the segmentation accuracy is affected by image noise and the segmentation algorithms used, inevitably resulting in less reproducibility in the LV function quantification. We propose a heuristic scheme to forward project (re-project) the short axis (SA) slices of gated blood pool SPECT to generate a quasi-2D stack of images to improve the signal-to-noise ratio and simplify the LV segmentation process. Fourteen patients were randomly selected for this study that had standard in vitro Tc99m labeling of red blood cells. Patients were first imaged following a low-dose (8 mCi) injection using a cardiac dedicated SPECT camera and were then imaged following a high-dose (25 mCi) injection using a conventional planar camera. Left anterior oblique (LAO) views were generated from 3-D SPECT using GE multi-gated software and SA reprojections were created via a new Yale method. Six different quantification methods were compared for quantitative accuracy of EDV, LVEF and PFR assessments. The new method that incorporated the SA reprojection scheme was most comparable to results from gold standard high-dose ERNA method in terms of EDV quantification. The new method allowed for a greater than 67% dose reduction from 25 mCi to 8 mCi without compromising the accuracy of EDV, LVEF and PFR quantification.

Keywords: SPECT, ERNA Quantification
Poster panel: 328

Poster Number:

Metal Artifact Reduction by Mapping MR into CT Information (#1213)

T. Shevaruangroj1, S. Auethavekiata1, V. Sa-ing2

1 Chulalongkorn University , Department of Electrical Engineering , Bangkok, Thailand
2 Rangsit University, Department of Radiological Technology , Pathum Thani, Thailand


One of the fundamental noise in Computed Tomography (CT) imaging is metal artifact. The artifact leads to the information loss. Magnetic Resonance (MR) imaging is not affected by this artifact. In this paper, we propose using the information from MR images to reproduces the information lost due to dental filling in CT images. First, we register the CT and MR images of the same patient. Then, we classify the skull into 3 regions: teeth, tissue near the teeth and others. For teeth and tissue near the teeth, we develop the mapping function between the intensity in the MR image to the mean intensity in the CT image. The pixels corrupted by the metal artifact are detected and their intensity are changed according to the mapping function with its MR intensity as the input. The proposed method significantly removed metal artifacts and could be used to improve the diagnosis and planning accuracy.

Keywords: Metal artifact reduction, Mapping function
Poster panel: 331

Poster Number:

Count Rate Corrections for the Plant Dedicated PET System phenoPET (#1342)

C. Hinz1, 2, S. Jahnke1, R. Metzner1, D. Pflugfelder1, J. Scheins3, M. Streun4, U. Pietrzyk3, 2

1 Forschungszentrum Juelich GmbH, Institut of Bio- and Geosciences: Plant Sciences (IBG-2), Juelich, North Rhine-Westphalia, Germany
2 University of Wuppertal, School of Mathematics and Natural Sciences, Wuppertal, North Rhine-Westphalia, Germany
3 Forschungszentrum Juelich GmbH, Institute of Neuroscience and Medicine: Medical Imaging Physics (INM-4), Juelich, North Rhine-Westphalia, Germany
4 Forschungszentrum Juelich GmbH, Central Institute for Engineering, Electronics and Analytics: Electronic Systems (ZEA-2), Juelich, North Rhine-Westphalia, Germany


PET imaging is widely used to perform clinical and preclinical studies. A rather new area is the field of plant research. PET allows the dynamic imaging of 3D carbon transport and allocation within plants shoot and root systems of plants. Here, a new system called phenoPET was developed within the German Plant Phenotyping Network (DPPN) using Digital Photon Counters (DPC) by Philips (digital silicon photo multipliers). Those are organized in horizontal rings forming a field of view (FOV) of 180mm diameter and 200mm height. The detected hits are arranged in time frames, which are written to disc via USB 3.0 with a maximum transfer rate of 300MB/s. Preprocessing and coincidence sorting are done offline. To enable high dynamic range phenoPET can drop frames which are not saved on disc when the transfer rate comes to a limit. To compare carbon transport of different plants quantitative images are required. A preliminary normalization and random correction are implemented. In this context count rate correction and dead time correction are mandatory to avoid quantitative bias. In a decay experiment the dead time of the DPCs reduces the rate of singles by up to 10% at an radioactivity of about 40MBq placed close to the center of the FOV. At this activity half of the frames are dropped leaving a single rate of about 6×106cps. Further count rate correction factors for coincidences due to missing frames are presented.

Keywords: Positron Emission Tomography, dynamic range, dSiPM, plant imaging
Poster panel: 334

Poster Number:

Accuracy of low dose and diagnostic CT image registration of bronchial tree for virtual bronchoscopy (#1592)

G. Opposits1, M. Nagy2, 1, C. Aranyi1, M. EMRI1, L. Balkay1

1 University of Debrecen, Faculty of Medicine / Medical Imaging Department / Division of Nuclear Medicine, Debrecen, Hungary
2 University of Debrecen, Faculty of Medicine / Medical Imaging Department / Department of Biomedical Laboratory and Imaging Science, Debrecen, Hungary


The aim of this study is to support medical experts to be able to make an order among large number of automatic registration. The experts could tackle with the most problematic cases due to the inaccuracy of automatic registration procedure in the vicinity of the bronchus to help virtual bronchoscopy (VB) systems. Functional images (e.g. PET) can be projected on the relevant part of the organ that is examined in VB systems. We collected cases where the difference between the time of low-dose (ldCT) and diagnostic CT (hdCT) was less than one year. Altogether 22 anonymous ldCT and hdCT studies were selected in this study. Based on the literature, a potential candidate for image registration was elastix. We applied a specific in-house developed application for image preprocessing, before the elastix registration. We tried to resolve the goodness of the entire registration process by visual judgment combining it with special numerical features. Numerical data include the marginal entropies joint entropy, mutual information, standardized mutual information, Kullback-Leibler entropy, cross correlation, L1norm, L2norm square. We found satisfying correlation between mutual information and visual judgment in the close vicintiy of airtree. We calculated confident intervals for MI of acceptable and rejected registrations about the mean values of it by bootstraping method: [0.27, 0.36], [0.13, 0.19].

Keywords: diagnostic, computed tomography, bronchoscopy, image registration, image segmentation
Poster panel: 337

Poster Number:

Single-Energy Single-Scan Material Decomposition for Contrast-Enhanced Breast Computed Tomography (#1761)

B. Kim1, H. Jeon2, H. K. Kim1, 3

1 Pusan National University, School of Mechanical Engineering, Busan, Republic of Korea
2 Pusan National University Yangsan Hospital, Department of Radiation Oncology, Yangsan-si, Republic of Korea
3 Pusan National University, Center for Advanced Medical Engineering Research, Busan, Republic of Korea


This paper describes phantom imaging studies to investigate the feasibility of a single-energy single-scan (SESS) material decomposition (MD) method for the contrast-enhanced breast computed tomography (BCT). While the conventional MD method requires dual-energy scan measurements, the investigating method requires a single scan measurement and X-ray penetration length data. The X-ray penetration length data is estimated from the conventional BCT image using the ray-tracing technique. We perform experimental simulations with the SESS-MD approach and assess its performance by measuring the contrast and noise, or the contrast-to-noise ratio (CNR). The performance is compared with the conventional CT and dual-energy CT (DE-CT) methods. In terms of contrast, the SESS and DE-CT outperform the conventional CT. We observe that the CNR performance of SESS is almost comparable to that of the conventional CT. On the other hand, the DE-CT shows the worst CNR performance because of the increased noise due to the subtraction operation. The performance of SESS will be further quantified for various geometries and operation parameters in comparisons with those of conventional methods.

Keywords: breast computed tomography, material decomposition, contrast agent, breast cancer
Poster panel: 340

Poster Number:

ML background scale factors estimation for prompt gamma tracers in PETCT imaging (#1798)

H. Bal1, V. Y. Panin1, M. Aykac1, M. Conti1

1 Siemens Medical Solutions USA, Inc, Molecular Imaging, Knoxville, Tennessee, United States of America


Correction of prompt gamma distribution and scatter due to true coincidences is required to obtain quantitative estimates for PETCT imaging with prompt gamma tracers. Currently, this background is estimated as a linear combination of single scatter simulation (SSS) and prompt gamma estimate (PGE) each scaled using tail fitting approach (TFBS). In this work we extend the maximum likelihood (ML) scatter scaling approach to compute scaling factors for both background components along with the activity image which is referred to as ML background scaling method (MLBS). An ML based solution with nested loop for PGE and SSS scaling factors estimation was obtained similar to that proposed by Panin. In order to validate this approach, experimental phantom studies with a uniform cylinder and image quality phantom filled with I-124 were performed. In addition, a preliminary assessment of visual image quality in two clinical cardiac Rb-82 studies for an average and a large patient was performed. Scatter fractions for both approaches were estimated for all studies. Uniformity in axial profiles for the uniform cylinder and the hot sphere contrast for the image quality phantom were used as figures of merit. Based on the uniform cylinder phantom, the number of iterations for estimating scale factors was selected to be 25 for each 6 image updates. For the experimental phantom studies, MLBS approach produces images visually similar to those with TFBS approach. However, reconstructed images for the uniform cylinder showed more axial uniformity with MLBS approach compared to TFBS approach. In the case of cardiac Rb-82 studies, MLBS and TFBS approach gave similar activity images for the average patient. However for the large patient, some background over-correction was observed with TFBS but not MLBS approach. ML background scale factor estimation approach performed equivalent to the current TFBS approach but has the potential to improve image quality in the case of large patients.

Keywords: Scatter scaling, Prompt gamma, PETCT
Poster panel: 343

Poster Number:

A Companion Source Based Detector Dead-Time Correction for Single Photon Emission Computed Tomography  (#1946)

X. Ding1, M. Bhattacharya1

1 Siemens Medical Solutions, HC DI Molecular Imaging , Hoffman Estates, Illinois, United States of America


Abstract—In many clinical applications of Single Photon Emission Computed Tomography (SPECT), such as scanning patients for dosimetry following the administration of therapeutic doses of radio-pharmaceuticals, detector count rates are rather high necessitating accurate detector dead-time correction in order to generate accurate quantitative measurements for the radio-pharmaceutical distributions within a body. Current SPECT dead-time correction methodologies typically need to modify the detector hardware and the corresponding front-end software that could be expensive. In this paper, we describe a novel dead-time correction method for SPECT detectors without altering any detector imaging pipeline. Our proposed method is based on using companion point sources. Before performing a required clinical SPECT tomographic scan, one first prepares two point sources using the same radio-isotope, and mounts them on the corner locations of two detectors, one point source per detector. One then performs a blank scan without the patient and the actual scan with the patient. In both scans, the companion point sources are imaged frame by frame with different count rates. We developed an algorithm and the corresponding software platform that automatically locate the position of the companion point source in each frame, and derive a dead-time correction coefficient per frame and per energy photo-peak. For a 6.3L uniformity cylindrical phantom filled with 107 mCi of Lu177, with the proposed dead-time corrections, the error rate of activity concentration between the reconstructed image and the truth is 1.7%, while without any dead-time corrections, the error rate would be 5%. Note that the error rate in the reconstructed activity concentration grows with increasing activity concentration. Note also that the method is independent of the particularities of detector front end signal processing and in essence is a universal method independent of the radio-pharmaceutical being imaged.

Keywords: Dead -time corretion, Qualitative SPECT, Companion Source
Poster panel: 346

Poster Number:

Attenuation correction in a combined, single-gantry breast PET-Tomosynthesis scanner (#1642)

S. Krishnamoorthy1, T. Vent1, B. Barufaldi1, A. D. A. Maidment1, J. S. Karp1, 2, S. Surti1

1 University of Pennsylvania, Dept. of Radiology, Philadelphia, Pennsylvania, United States of America
2 University of Pennsylvania, Dept. of Physics and Astronomy, Philadelphia, Pennsylvania, United States of America


We are currently building a breast PET-Tomosynthesis scanner. The PET & Tomosynthesis scanners will be integrated onto a single gantry, providing spatially co-registered 3D PET-Tomosynthesis images. Clinical studies have demonstrated the benefits of tomosynthesis over mammography, especially in identifying lesions within dense breasts. We hypothesize that the tomosynthesis scanner can aid in improving PET quantitation by improving attenuation correction (AC) based on identification of the tissue-types present in the breast tomosynthesis image. The PET detector consists of a 32 x 32 array of 1.5x1.5x15 mm3 LYSO crystals. The PET scanner utilizes two detector heads separated by 9 cm, with each detector head having a 4x2 arrangement of the PET detectors. GATE simulations were performed with an anthropomorphic breast phantom that was generated using the Virtual Clinical Trial (VCT) framework that allows building synthetic breast models based on clinical breast scans. The initial study was performed with a compressed breast that measured 6.8 x 12.6 x 6.15 cm3 in size. While the VCT environment permits the definition of a variety of tissues, our study limited the tissue classification to air, breast tissue, adipose & skin. An FDG-avid 5 mm diameter lesion with specified uptake ratio was also simulated at the center of the breast phantom. GATE simulation data was converted to list-mode format & image reconstruction performed with a blob-based iterative algorithm using a 300ps TOF kernel. Reconstruction either excluded AC, or, utilized tissue segmentation & classification as used in the breast model to perform AC. Our initial studies show that AC provides a better estimate of the true tracer uptake. Work is underway to utilize the reconstructed DBT image for tissue segmentation, classification & its subsequent use in performing AC for PET. We will also vary the breast density and evaluate accuracy & precision of the different tissue-classification techniques to perform AC.

Keywords: Breast PET, Time-of-flight, Attenuation Correction, PET-Tomosynthesis, Positron Emission Tomography
Poster panel: 349

Poster Number:

Post-Reconstruction Positron Range Correction for Preclinical and Clinical PET Imaging (#2163)

J. Cal-Gonzalez1, J. L. Herraiz2, 3, J. M. Udias2, 3, M. Conti4, C. Kuntner5, M. Blaickner5, T. Beyer1

1 Medical University of Vienna, Center for Medical Physics and Biomedical Engineering, Vienna, Wien, Austria
2 Universidad Complutense de Madrid, Grupo de Física Nuclear and UPARCOS, Madrid, Spain
3 Instituto de Investigación Sanitaria del Hospital Clínico San Carlos (IdISSC), Madrid, Spain
4 Siemens Medical Solutions USA, Inc., Molecular Imaging, Knoxville, Tennessee, United States of America
5 AIT Austrian Institute of Technology GmbH, Biomedical Systems, Center for Health & Bioresources, Seibersdorf, Austria


Positron range (PR) is a significant factor that limits PET image resolution. Its effect is especially relevant in several radionuclides currently used in clinical and preclinical studies such as 82Rb, 124I and 68Ga. The objective of this work is to implement a practical accurate positron range correction (PRC), which can be applied directly to the reconstructed images, without the need to access the acquired data or to modify the image reconstruction software.

The proposed post-reconstruction positron range correction (post-PRC) method is an iterative Lucy-Richardson algorithm based on a spatially-variant PR blurring kernel derived from an anatomical image (CT or MR) for the specific radionuclide. The method was validated with two simulated datasets (Argus PET/CT preclinical system), where the no-PRC and a reconstruction-based PRC (IR-PRC) were also available. After validation, the method was applied to images of a micro-hollow sphere phantom filled with 68Ga acquired with a preclinical Focus-220 PET system; and to images of a NEMA Image Quality (IQ) phantom, filled with 124I acquired in a clinical Biograph mCT PET/CT system.  

The post-PRC method provided similar images to the IR-PRC for the simulated datasets, with only a slightly smaller resolution recovery. Both post-PRC and IR-PRC methods provided artifact-free images in the simulated datasets with hot sources located close to air cavities. The application of the post-PRC in the experimental phantom acquisitions resulted in significant resolution recovery both for preclinical (~50% increase) and clinical acquisitions (~25% increase). 

In summary, we have developed a practical post-PRC methodology, which provides accurate results and can be easily applied to any preclinical or clinical PET systems. 

Keywords: positron range correction, post-reconstruction, PET imaging
Poster panel: 352

Poster Number:

Correction of Bulk Motion Using Real-time MRI for PET Motion Correction in Hybrid PET/MRI (#2273)

L. J. Frohwein1, F. Büther1, 2, K. P. Schäfers1

1 University of Münster, European Institute for Molecular Imaging (EIMI), Münster, North Rhine-Westphalia, Germany
2 University Hospital of Münster, Department of Nuclear Medicine, Münster, North Rhine-Westphalia, Germany


Patient motion during PET scans is well-known to degrade image quality. Hybrid PET/MRI scanners are able to acquire motion information with the MRI to complement the PET reconstruction. There are several methods to incorporate information about respiratory and cardiac motion into the PET image reconstruction using motion models and gating. However, most methods fail when bulk body motion or irregular, non-predictable motion patterns are present during the scan. We present an MRI-based continuous, real-time motion tracking in which respiratory and cardiac motion as well as bulk motion are taken into account. MRI data is acquired throughout the whole PET scan session with a sampling frequency of 700 ms/volume (with 20 slices, 35 ms/slice). Initially, the data is analyzed towards bulk motion before a slice resorting is created to ensure a temporal volume consistency. Results show that bulk motion can be detected with the method and corrected using rigid transformation. The subsequent slice resorting leads then to temporally consistent volumes that can be used to create motion vectors which can be in turn applied to the PET data.

Keywords: motion correction, PET/MRI, bulk motion, real-time MRI
Poster panel: 355

Poster Number:

Retrospective Motion Correction in Chemical Exchange Saturation Transfer (CEST) MRI Using Gradient-based Motion Correction (GradMC) Algorithm (#2344)

D. H. Lee1, D. W. Lee2, J. I. Kwon3, S. I. Lee3, D. C. Woo3

1 The University of Sydney, Faculty of Health Sciences and Brain and Mind Centre, Sydney, Australia
2 Asan Medical Center, Center for Bioimaging of New Drug Development and MR Core Laboratory, Asan Institute for Life Sciences, Seoul, Republic of Korea
3 Asan Medical Center, MR Core Laboratory, Asan Institute for Life Sciences, Seoul, Republic of Korea


Chemical exchange saturation transfer (CEST) has been used as a molecular magnetic resonance imaging (MRI) technique to indirectly detect the metabolite content based on exchange-related properties. In particular, the glutamate CEST (GluCEST) is a newly developed method to image parenchymal glutamate in the brain by measuring the exchange of glutamate amine protons with bulk water. However, as the main disadvantage, CEST requires a relatively long scan time to collect the whole voxel analysis data and z-spectrum while varying the resonance frequency around the water. In this study, we applied a retrospective motion correction approach using gradient-based motion correction (GradMC) algorithm to CEST data in order to investigate the feasibility of motion correction. Motion corrupted CEST data were used to induce arbitrary head motion inside the head coil through pre-clinical epilepsy rat models. The epileptic seizure was induced in a rat by 15 mg/kg intraperitoneal injection of kainic acid (KA). CEST data were obtained using a 7.0T Bruker MRI scanner before and after (3 hours) KA injection. GradMC algorithm iteratively searches for the motion trajectory yielding the sharpest image as measured by the entropy of spatial gradients. The vast space of motion parameters is efficiently explored by gradient-based optimization with a convergence guarantee. Our results clearly showed that the magnetization transfer ratio asymmetry (MTRasym) curves after correcting the motion showed an improved result with significantly reduced extreme signal changes and large standard deviations than the results from uncorrected data. To compare the GluCEST signal values at 3.0 ppm, there was only statistically significant difference in the result of motion free data and the result of motion corrected data (P = 0.038; Wilcoxon signed rank test). Our results also showed that the GradMC can be used to efficiently correct for motion occurring during CEST imaging.

Poster panel: 358

Poster Number:

3D Segmentation of Artery 18F-FDG PET/CT Images in Elderly utilizing Affinity Propagation (#2616)

M. Y. Mokeddem1, M. Al-enezi1, R. - A. Abdo1, M. Bentourkia1

1 Université de Sherbrooke, Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada


We present in this study a semi automatic procedure to segment PET artery images in elderly subjects with atherosclerosis inflammation. The inflammation in the artery is a precursor of atheromatous plaque detachment and obstruction of blood vessels in the brain or the heart. We used a bimodality computed tomography (CT) and dynamic positron emission tomography (PET) with 18F-FDG which is a biomarker used in arteries presenting high metabolic activities. CT imaging allows to detect the calcifications in the arteries, while PET imaging allows to detect the metabolically active inflammation. The goal of the artery image segmentation was to correlate calcification on CT images and artery metabolism on PET images. First, all the artery images were delineated in each slice on the non-enhanced CT images with reference to anatomic atlases, and we used the active contours to determine the region of interest (ROI) on the corresponding PET artery images at early time frames. All the artery images on PET were corrected for PVE then each slice of PET images were segmented with the algorithm of affinity propagation (AP). The 3D reconstruction of the segmented arteries allowed to show the presence of inflammation, and also the low uptake of 18F-FDG in artery segments corresponding to large area of calcification. In conclusion, the combination of CT and PET imaging with appropriate 3D image segmentation can be of great importance in locating vulnerable atherosclerotic plaques.

Keywords: PET imaging, Atherosclerosis, Segmentation, CT imaging, Arteries
Poster panel: 361

Poster Number:

A method to estimate motion vector for rigid-body motion from  PET without motion tracking device (#2782)

I. Hong1, Z. Burbar1

1 Siemens Medical Solutions USA, Inc., Molecular Imaging, Knoxville, Tennessee, United States of America


Motion in PET studies degrades the image quality and introduces bias which is critical for high 
resolution scanners. There are many publications related to motion correction in PET. Most of 
these methods rely on external devices to track the motion and reg ister it to the listmode data. 
This  paper  describes  how  to  extract  patient  subject’s  rigid-body  head  motion  from  only  the 
listmode data. Starting with the listmode data, the data is histogrammed into small time frames 
(e.g.  1  sec)  into  3D  sinograms.  These  3D  sinograms  are  then  reduced  into  2D  sinogram  by 
summing  all  the  axial  planes  including  the  oblique  segments.  The  sinogram  is  then 
reconstructed using 2D iterative reconstruction such as 2D OSEM. The 3D normalization file is 
also  summed  the  same  way  as  the  3D  sinogram.  This  norm  is  used  for  the  2D  iterative 
reconstruction. Each output image is then registered to the first reconstructed image to generate 
X and Y translation and a rotating angle  by using mutual information. The results are then the 
patient  head motion in X and Y translation and the rotating angle in respect to time. This  is a 
simple indication of the patient motion that will assist the user to extract 3D motion vectors at 
the time of the motion that is detected with this method.

Keywords: Brain Study, Motion Correction, Data Driven Gating
Poster panel: 364

Poster Number:

Delineation techniques of tumor hypoxia volume with 18F-FMISO PET imaging (#1900)

R. - A. Abdo1, F. Lamare2, M. Allard2, P. Fernandez2, M. Bentourkia1

1 Université de Sherbrooke, Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada
2 Université de Bordeaux2- EPHE, Service de Médecine Nucléaire, Bordeaux, France


18F-fluoromisonidazole (18F-FMISO) is the most used PET hypoxia biomarker. Uptake of 18F-FMISO by tumors provides a visual and a quantitative map of hypoxia, which can be used for dose escalation in radiotherapy. Several techniques have been developed to quantify and delineate the tumor hypoxia volumes such as tumor-to-blood ratio (TBR), tumor-to-normal tissue ratio (TNR) and the modeling methods. This study aims to report quantitative analyzes in delineating hypoxia in tumors. We scanned 9 subjects with 18F-FMISO PET/CT and MRI. All subjects were diagnosed with glioblastoma tumor. Scanning protocols involved a 15 or 30 min dynamic scan followed by 10 min static scan at 2 h, 3 h and 4 h in order to allow accumulation of 18F-FMISO in the tumors. The quantitative analyzes were performed with spectral analysis (SA) and by gradation with TBR and TNR. TBR and TNR images were obtained by normalizing the measured images with respect to the concentration of 18F-FMISO in blood and in normal tissue, respectively with multiple threshold levels of 1.2, 1.4, 2 and 2.5. A similarity index (SI) approach was chosen to quantitatively compare the TBR, TNR and SA images. The results showed that the TBR technique produced less spatially constrained volumes around the tumor than TNR with the same threshold level. SA images show the decomposed components of tumor as accumulative regions in the center of the tumor while the perfused regions were seen on the peripheries. The hypoxia defined with SA was more accurate since it is not sensitive to input function artefacts such as the partial volume. We found some similarities at specific thresholding levels and imaging times between TBR, TNR and SA. At 3 h, TBR at a level of 2.5 was close to TNR at 1.4 (SI > 95%), and TNR at 2 h and 3 h with a level of 1.2 presented a high similarity (SI > 85%). Despite these similarities, more interventions are expected for proper decisions.

Keywords: Tumors, Glioblastoma, PET imaging, Spectral analysis, tumor-to-blood ratio
Poster panel: 367

Poster Number:

De-noising and DA release: effect of denoising on the ability to identify voxel level neurophysiological response (#2605)

C. W. J. Bevington1, J. - C. Cheng1, 2, I. S. Klyuzhin3, V. Sossi1

1 University of British Columbia, Department of Physics and Astronomy, Vancouver, British Columbia, Canada
2 University of British Columbia, Pacific Parkinson's Research Centre, Vancouver, British Columbia, Canada
3 University of British Columbia, Department of Medicine, Division of Neurology, Vancouver, British Columbia, Canada


An important aspect in identifying spatiotemporal neurophysiological response in PET brain imaging is the ability of data reconstruction and de-noising to increase the quality of the data without sacrificing quantitative accuracy. Previous work has shown accurate modeling of time-varying neurotransmitter levels with the linear parametric neurotransmitter PET (lp-ntPET) model on the voxel level, in dynamic PET data reconstructed with Ordered Subsets Expectation Maximization (OSEM) and HYPR post-processing. Here we expand on this work by testing lp-ntPET on a simulated DA release detection protocol consisting of 11C-Raclopride (RAC) scans ‘acquired’ after bolus injection with intervention-induced dopamine (DA) release starting at 35 min after injection. Simulation parameters where obtained from real patient data. The simulated data were reconstructed with a variety of reconstruction and post-processing techniques. Unlike previous methods, Poisson noise is added after forward-projecting the noiseless data to better simulate the imaging process. We find that for a 15% PET-estimated DA release (percentage decrease in RAC BPND ), a denoised reconstruction technique (HYPR-AU-OSEM) combined with HYPR post-processing produces a ~25% higher recovery of areas involved in dopamine release (sensitivity) compared to previously explored methods (OSEM with HYPR post-processing). Increasing the kernel size in HYPR-AU-OSEM was found to increase recovery of the response, but also requires more computational time. As well, recovery of the strength of the DA release was more accurate and precise with this most sensitive reconstruction technique. Future work will involve simulating multiple realizations, different simulated peak DA release levels, assessment of a fully-4D de-noising reconstruction and testing with patient data.

Poster panel: 370

Poster Number:

Initial Comparison of Two Software Packages for Processing SPECT MBF Studies (#2718)

R. G. Wells1, A. Lakhani1, B. F. Marvin1, R. A. deKemp1, T. D. Ruddy1

1 University of Ottawa Heart Institute (UOHI), Cardiac Imaging, Ottawa, Ontario, Canada


Dynamic SPECT with stationary cardiac cameras allows measurement of myocardial blood flow (MBF) and flow reserve (MFR), but clinical tools to perform kinetic analysis with SPECT are limited.  We compare the MBF values calculated using two different software packages.

Dynamic SPECT datasets were acquired using a multi-pinhole cardiac SPECT camera and a 1-day rest/stress tetrofosmin protocol (350 MBq at rest, 1100 MBq at stress).  Listmode data from 20 patients were rebinned into 9 x 10 s, 6 x 15 s, and 4 x 120 s time frames and reconstructed using a vendor-supplied MAP-EM algorithm.  Tracer kinetic analysis was done with a 1-tissue-compartment model using 4DM and FlowQuant (FQ) software. Studies were compared with and without manual motion correction (MC) for inter-frame patient motion. A second observer repeated the MC to assess inter-user variability. Data from a 2nd set of 21 patients were processed by a third user and similarly compared. 

 Without MC, the correlation between 4DM and FQ was r=0.77. The 4DM values averaged 1.87 times the FQ values.   Comparing the MFR values, the correlation was r=0.64 and the average difference was 0.4 ± 0.6.  When MC is applied, the correlation in MBF between 4DM and FQ was r=0.82 and r=0.77 for the two users. The slope of the linear fit was 1.74 and 1.56 respectively. The inter-user correlation was r=0.79.  The correlation between 4DM and FQ for MFR with MC was r=0.71 and r=0.44 for the two users. The inter-user MFR correlation was r=0.86.  A 3rd user on a 2nd set of 21 patients found a MBF correlation of r=0.78 (no MC) and r=0.58 (MC) with linear regression slopes of 1.64 (no MC) and 0.66 (MC). The MFR correlation was r=0.62 (no MC) and r=0.69 (MC). The mean MFR difference was 0.1 ± 1.1 (no MC) and -0.5 ± 0.8 (MC). Large changes in the regression line slopes between the 3 users show MC has a strong user-dependence.

Correlation between 4DM MBF and FQ MBF is moderate (r=0.7 – 0.8) and 4DM MBF is 1.5-1.7 times FQ MBF with no MC.

Keywords: SPECT, cardiac, Myocardial Blood Flow
Poster panel: 373

Poster Number:

Non-invasive parametric mapping of binding in mouse brain-PET studies: a validation (#2930)

C. J. Wimberley1, D. - L. Nguyen1, C. Truillet1, Z. Gullan2, M. - A. Peyronneau1, M. Tonietto3, R. Boisgard1, S. Chalon2, V. Bouilleret1, I. Buvat1

1 Inserm/CEA/University Paris Sud, Laboratoire Imagerie Moleculaire In Vivo (IMIV), Orsay, France
2 University of Tours/INSERM, U 930 Imagerie et Cerveau, Tours, France
3 L'hôpital de la Pitié-Salpêtrière, L'Institut du Cerveau et de la Moelle Épinière, Paris, France


TSPO PET imaging can follow neuroinflammation associated with neurodegenerative disorders using [18F]-DPA-714, but there are quantification challenges because longitudinal arterial sampling in mice is nearly impossible. Due to the ubiquitous expression of TSPO, no reference region can be defined. The aim of this study was to use a previously described method for the extraction and use of an image derived input function (IDIF) using a factor analysis (FA) approach which was input to a voxel wise Logan plot to create parametric maps of binding. These maps were then validated against quantified autoradiography within the same animal.

The model was induced by injection of kainic acid into the right hippocampus of adult male C57/Bl6 mice (n=4). A dynamic [18F]DPA-714 PET/CT (60 min) was performed 1 month and 6 months post KA injection. FA was applied to all images using 4 factors to extract the IDIF. The normalised curve was used in a voxel wise Logan plot to estimate the total volume of distribution, VT. Each animal was sacrificed after the final scan and the brain underwent autoradiography with 3H-DPA-714 (4 slices starting at the hippocampus). Regions of interest were manually drawn over brain regions on each slice and equivalent regions were drawn on the VT map. The average regional binding values were extracted and correlations were generated for each mouse.

Each of the four mice showed strong correlations between the average VT extracted from the ROIs and the autoradiography: 1 month: r2 = 0.62 and 6 months r2 = 0.75, 0.68 and 0.77. The coefficient of determination for all regions within all mice together was 0.59.

This study showed strong regional correlations between VT maps and autoradiography, and the maps showed strong visual similarities. These results validate the use of an IDIF extracted using a FA approach to produce parametric maps. This method will be useful for longitudinal studies within rodents and for studies using whole body PET.


Keywords: parametric mapping, image derived input function, mouse, brain, TSPO
Poster panel: 376

Poster Number:

Longitudinal measurements of multiparametric MRI: A phantom study (#1500)

Y. F. Wang1, 2, G. Liney2, 3, R. Rai2, 3, L. Lolloway2, 3, A. Haworth1, 2

1 University of Sydney, Institute of Medical Physics, Sydney, Australia
2 Ingham Institute for Applied Medical Research, Sydney, Australia
3 Liverpool and Macarthur Cancer Therapy Centre, Sydney, Australia


Introduction: Quantitative imaging using multiparametric MRI (mpMRI) has been extensively studied to provide potential biomarkers for cancer diagnosis and treatment response. Differences between MRI systems (inter-scanner), and drifts of parameters within the same system (intra-scanner) contribute to uncertainties in the quantitative data which must be evaluated to rationalise data from longitudinal and multi-centre studies. The goal of this study is to investigate the variability in mpMRI parameters including apparent diffusion coefficient (ADC), T2, T1, R2* using an inhouse developed multicompartment phantom.

Methods: A multicompartment phantom was developed consisting of ethanol, gelatine, and gadolinium-doped (Gd) water to simulate a suitable range of in vivo tissue parameters. The phantom was scanned on a daily basis on a 3.0-T MRI scanner to generate ADC, T1, T2, and R2* maps using standard clinical sequences. ADC, T1, R2* maps were acquired from the scanner and T2 maps were calculated by fitting the mono-exponential decay signal function. The parametric values for each sample were extracted from a region of interest, and the coefficient of variance (COV) was calculated to evaluate the variability of the values in the daily measurements.

Results:. Minimum variation was found in the T2 maps with COV ranging from 0.48% to 3.98%, while maximum variation was found in the R2* maps, with COV ranging from 1.06% to 20.55%. The 4.3mM Gd-doped water had lowest COV and gelatin sample had highest COV across all parametric maps.


A preliminary study with the inhouse mpMRI phantom has been used to assess reproducibility of ADC, T1, T2, and R2* maps on a clinical scanner. Based on the stability of the measurements, it was found that R2* measurements have the lowest reproducibility and T2 measurements have the highest reproducibility. Future work will focus on quantifying the longitudinal variability of mpMRI parameters across multi-centres.

Keywords: phantom, multiparametric MRI, repeatability, reproducibility, longitudinal imaging
Poster panel: 379

Poster Number:

Evaluation of photon counting spectral mammography for classification of breast microcalcifications (#1670)

H. Kim1, D. Kim2, M. Lee2, H. - J. Kim1, 2

1 Yonsei University, Department of Radiological Science, College of Health Science, Wonju, Republic of Korea
2 Yonsei University, Department of Radiation Convergence Engineering, College of Health Science, Wonju, Republic of Korea


 The purpose of this study was to evaluate the feasibility of spectral mammography using the dual-energy method to non-invasively distinguish between type I (calcium oxalate, CO) and type II (calcium hydroxyapatite, HA) microcalcifications. Two types of microcalcifications are difficult to distinguish due to a similar linear attenuation coefficient. In addition, to investigate the detection efficiency of microcalcifications, we compared two detector materials (Cadmium Zinc Telluride (CZT) and silicon), which are widely used as photon counting systems.
 In this study, photon counting spectral mammography system was simulated using Geant4 Application for Tomographic Emission (GATE) simulation tools. For the comparison, two detector materials (CZT and silicon) were used in the same geometry. The thickness of the breast phantom was 3 cm and microcalcifications of various sizes ranging from 130 μm to 550 μm were embedded into the breast phantom. Microcalcifications were classified as being calcium hydroxyapatite or calcium oxalate based on score calculation with the dual energy images. 
 As a result, the measured CNR of CZT-based spectral mammography was higher than that of the silicon-based system. Silicon-based images showed more noise than CZT materials. The high atomic number of the materials gives a high quantum efficiency in comparison with silicon. In addition, the classification of microcalcifications showed better performance in CZT detectors. Due to the large noise characteristic, the results obtained from the silicon detector showed that there was a variation depending on the size of each calcification, and a switching part occurred between the two curves (HA and CO).
 This study demonstrated the feasibility of photon counting spectral mammography for classification of breast microcalcifications. These results are expected to potentially improve the efficiency of early breast cancer diagnosis.

Keywords: photon counting spectral mammography, breast microcalcifications
Poster panel: 382

Poster Number:

Investigation of Image Quality on Short-Scan CBCT Reconstruction (#2042)

C. Thanasupsombat1, S. S. Thongvigitmanee1, S. Aootaphao1, P. Thajchayapong2

1 National Electronics and Computer Technology Center, X-Ray CT and Medical Imaging Laboratory, Biomedical Electronics and Systems Research Unit, Pathumthani, Thailand
2 National Science and Technology Development Agency, Pathumthani, Thailand


In the cone-beam computed tomography system, the short scan approach was proposed to reduce the scan time, thus minimizing the motion artifacts and the patient’s radiation doses. Since the sinogram from the short scan contained redundant data causing new artifacts in the reconstructed images, a weighting function was applied to the FDK reconstruction algorithm to reduce the effects of the redundant data. In this work, we applied the so-called Parker weights and investigated the image quality of the reconstructed images. The results showed that CT numbers and uniformity acquired from the short scan were comparable with that of the full rotation scan, but image noise was increased. Furthermore, different initial angles of the short scan caused different artifacts on the reconstructed image and slightly affected the spatial resolution; however, the overall image quality was not significantly different.

Keywords: cone-beam CT, short scan, image quality, parker weights
Poster panel: 385

Poster Number:

Effect of Incomplete Acquisitions on the Image of High Resolution Preclinical Scanners (#2235)

P. Galve1, J. L. Herraiz1, 2, A. Lopez-Montes1, C. Gutierrez Fernandez3, J. M. Udias1, 2

1 Universidad Complutense de Madrid, Grupo de Fisica Nuclear and UPARCOS, Madrid, Spain
2 Instituto de Investigación Sanitaria del Hospital Clínico San Carlos (IdISSC), Madrid, Spain
3 SEDECAL Pre-clinical Division, Madrid, Spain


Positron Emission Tomography (PET) scanners are commonly made of many detector blocks that sample the field-of-view (FOV) with millions of lines-of-response (LOR). Sometimes some of the detector blocks may fail, or they could be intentionally removed to provide easier access to the animal or patient. In these situations, images obtained with basic reconstruction methods, like Filtered Back Projection (FBP) which assumes complete projection sampling, may contain artifacts. Iterative methods such as the Maximum Likelihood Expectation Maximization (MLEM) are more flexible and the impact of missing detectors on image quality will be low, specially for high-granularity scanners with a large number of detectors. We have chosen the Super Argus six-ring preclinical PET/CT scanner and studied with phantom acquisitions the impact  of removing a) up to half of its 144 detectors at arbitrary positions, and b) a whole section of the scanner, i.e. a pair of contiguous detectors from every ring. Image quality analysis following NEMA NU 4–2018 show only minor impact on MLEM images if less than 10 detectors are removed, 7% of the detectors for this scanner. On average, a mere 3% reduction in the recovery coefficient (RC) for the 1 mm rod and 1.6% noise increase was seen. On the other hand, the removal of a section of 12 detectors of the scanner does not significantly affect image quality in the central region (<1% loss in RC for the 1 mm rod and 4% noise increase), but artifacts may appear outside the center depending on the location of the section removed relative to the phantom. Regularization methods are being considered to improve this. In conclusion, the results show the robustness of MLEM respect to removal of a significant number of detector blocks.This could be used to design new PET scanner geometries that allow interacting with the patient while the data are being acquired, and to use real-time imaging to guide biopsy.

Keywords: Positron Emission Tomography, Image quality analysis, NEMA NU 4 - 2018, incomplete sampling
Poster panel: 388

Poster Number:

Accuracy of Attenuation Correction Strategy in Simultaneous PET/MR Evaluated with Simulated Lesions (#2768)

R. Laforest1, M. Savaikar1

1 Washington University, Medical School, Radiology, St.Louis, United States of America


Attenuation Correction in simultaneous PET/MR was originally implemented using the two-point DIXON which neglected the bones. More recent implementations of attenuation strategy in PET/MR involve either a fast UTE (or ZTE) MR sequence or a bone atlas which allows for inclusion of bones. However, soft-tissues are still segmented to fixed values to provide anatomical attenuation into four compartment identified as air, lungs, fat, water and for which bones are added as fifth compartment. Evaluation of the quantitative performance of these recently implemented attenuation methods is needed and this abstract presents an evaluation of the accuracy of  SUV (Standard Uptake Value) values from the choice of attenuation correction strategy. This work is done using a lesion insertion tool which is used to insert a spherical lesions into human body habitus of PET raw data acquired from PET/MR images to generate MonteCarlo based sinogram and reconstructed images.  The tool is presented and evaluated using digital phantoms. In addition, FDG PET data sets from cancer patients imaged by PET/MRI were used to study the impact of the choice of attenuation correction algorithm.  For these patients, a prior CT (Computed Tomography) scan was registered to the MR based attenuation correction images, scaled to PET attenuation coefficients, to provide a comparative CT based attenuation correction whihc was taken as the gold standard.  As expected, bony lesion SUVs are demonstrated to be underestimated by neglecting the bones in the attenuation correction. This bias is shown to be small bias when bones are included in the attenuation correction. However, soft-tissue lesions show a small dependency on the attenuation correction with overestimating SUVs when the bones are not included.   

Poster panel: 391

Poster Number:

First investigation of gadolinium-based nanoparticles for radio-sensitisation and enhanced imaging on the australian MRI-linac (#1083)

H. L. Byrne1, G. Le Duc2, F. Lux3, 2, O. Tillement3, 2, G. Liney4, T. Roberts4, 5, Z. Kuncic1

1 The University of Sydney, School of Physics, Sydney, Australia
2 NH Theraguix, F69100 Villeurbanne, France
3 Université Claude Bernard Lyon 1, 3. Institut Lumière-Matière, 69622 Villeurbanne, France
4 Ingham Institute for Applied Medical Research, Liverpool, Australia
5 Western Sydney University, School of Medicine, Sydney, Australia


High atomic number nanoparticles have the capacity to boost the therapeutic efficacy of radiotherapy by enhancing tumour-localised radiation damage. AGuIX is a gadolinium-based nanoparticle that has been shown to increase survival times compared to radiotherapy alone in pre-clinical animal studies. In addition to its high atomic number, gadolinium also provides positive contrast in magnetic resonance imaging (MRI) and is able to cross the blood-brain barrier where disrupted by a tumour. To this end, pre-clinical studies in a rat brain tumour model have demonstrated the synergistic effects of MRI-guidance and targeted radiotherapy using AGuIX. The first-in-man injection of AGuiX was successfully performed at Grenoble hospital in July 2016 within the phase I NANORAD clinical trial. 15 patients have been injected showing good drug tolerance, tumour targeting and contrast uptake and indications of clinical benefit according to the criteria that were used.

Here we report on the progress of the first study of AGuIX using an integrated MRI-linac. The Australian MRI-Linac is a dedicated research facility combining a linear accelerator for delivery of therapeutic radiation with a custom-designed 1.5 T magnet, allowing MR images to be acquired during treatment. We are replicating pre-clinical rat brain tumour model studies that were previously performed with separate MRI and radiotherapy. The MRI contrast provided by AGuIX not only aids delineation of the tumour target for verification and/or adaptation of the treatment plan, but can also monitor the uptake of gadolinium in the tumour to optimise the treatment delivery time to achieve radio-therapeutic gain from radio-enhancement effects. This work paves the way for a clinical trial of AGuIX in Australia.

Keywords: MRI, Nanoparticles, Radiosensitisation, megavoltage, Gadolinium
Poster panel: 394

Poster Number:

Validation of Geant4 Monte Carlo Toolkit Physics Models for Use in Heavy Ion Therapy (#1181)

A. Chacon2, 1, H. J. Rutherford2, 1, A. Mohammadi3, T. Yamaya3, G. Akamatsu3, Y. Iwao3, H. Tashima3, M. Nitta3, S. Takyu3, F. Nishikido3, A. Kitagawa3, T. Hofmann4, M. Pinto4, K. Parodi4, M. - C. Grégoire1, A. B. Rozenfeld2, S. Guatelli2, M. Safavi-Naeini1

1 Australian Nuclear Science and Technology Organisation, Human Health Research Theme, Lucas Heights, Australia
2 University of Wollongong, Centre for Medical Radiation Physics, Wollongong, Australia
3 National Institute of Radiological Sciences, National Institutes for Quantum and 13 Radiological Science and Technology, Inage Ward, Japan
4 Ludwig-Maximilians-Universität München, Medical Physics, München, Germany


Carbon and oxygen ion therapies are highly conformal radiotherapy modalities which can deliver a therapeutic radiation dose to a tumour while limiting significant damage to surrounding healthy tissue or organs. However, compared to standard photon therapy, treatment with carbon or oxygen ion beams is highly sensitive to positioning error due to the steep dose profile, which can result in under-dosing the tumour and/or overdosing healthy tissue. One method for verifying the dose distribution delivered to the patient is to image the distribution of short-lived positron-emitting fragmentation products generated as the beam passes through the patient, using an in-beam PET system. In order to develop and validate models relating the delivered dose to the PET image, it is necessary to have a reliable Monte Carlo simulation model which accurately reflects the physics of the interaction of carbon and oxygen ion beams with human tissue. Unfortunately, modelling nuclear interactions and the resulting secondary particle production is highly complex, because it involves high-energy interactions of a diverse range of nuclei, for which no fully validated models currently exist. In this work, we present the result of our experimental quantification of the production of positron-emitting nuclei during carbon and oxygen ion therapy across a range of homogeneous phantoms (water, polyethylene, poly(methyl methacrylate)) using a high resolution in-beam PET scanner with a resolution of 2.5 mm. Experimentally-estimated yields for each identified positron-emitting fragmentation product are compared with those predicted by a range of physics models available in the Geant4 Monte Carlo Toolkit. A comprehensive analysis of each physics model in comparison to the experimental results will be presented, including a recommendation of which model most accurately matches experimental results.

Keywords: Geant4, Heavy Ion Therapy, GATE, TOPAS, PET
Poster panel: 397

Poster Number:

Precise dose estimation by the use of luminescence imaging of water for proton therapy (#1260)

S. Yamamoto1, T. Yabe1, T. Toshito2, M. Yamaguchi3, N. Kawachi3, M. Komori1

1 Nagoya University, Graduate School of Medicine, Nagoya, Japan
2 Nagoya Proton Therapy Center, Nagoya, Japan
3 National Institutes for Quantum and Radiological Science (QST), Takasaki, Japan


Although the luminescence images of water during proton-beam irradiation using a cooled charge-coupled device (CCD) camera showed almost the same ranges of proton beams as those measured by an ionization chamber, the depth profiles showed lower Bragg peak intensities than those measured by an ionization chamber and a broad optical baseline was observed at deeper area than the Bragg peak. Also the luminescence intensity linearity for proton energy was not obvious yet. If we could solve and clarify these subjects, luminescence imaging of water will be used for proton therapy dosimetry. We hypothesize that the broad baseline signal originates from the light by the interaction of proton-induced prompt gamma photons with water. We corrected the measured depth profiles of the luminescence images by subtracting the simulated distributions of the produced light by the interactions of prompt gamma photons in water. We also calculated the luminescence intensity per proton for the luminescence images of water with different proton energies. With the correction, we successfully obtained the depth profiles that have almost identical distributions as the simulated dose distributions for protons. We also confirmed that the luminescence intensity of water per proton particle linearly increased with the proton energy. With these results, we confirm that the luminescence imaging of water can be used for the dose measurements of proton therapy.

Keywords: Proton, dose
Poster panel: 400

Poster Number:

Cerenkov Luminescence Imaging for the Heavy Ion Beam Range Verification - GATE Simulation Study (#1427)

H. G. Kang1, A. Mohammadi1, F. Nishikido1, C. Toramatsu1, 2, T. Yamaya1

1 National Institute of Radiological Sciences (NIRS) in National Institutes for Quantum and Radiological Science and Technology (QST), Department of Radiation Measurement and Dose Assessment, Chiba, Japan
2 Tokyo Women's University School of Medicine, Department of Radiation Oncology, Tokyo, Japan


Heavy ion therapy can deliver the optimal dose to a tumor while minimizing the unnecessary dose to normal tissues. Since the Bragg peak of a heavy ion beam is sharper than that of the proton beam, the beam quality control of the heavy ion beam is more important than the conventional proton beam. The aim of this study is to estimate the heavy ion beam depth in a water phantom by using Cerenkov luminescence imaging (CLI) with GATEv6.2 Monte Carlo simulation for beam quality control. A water phantom (10 × 10 × 60 mm3) was irradiated by heavy ion beams of carbon-12, carbon-11, and oxygen-15. The Cerenkov luminescence light emitted from the water phantom was detected by an optical system which consists of a lens (f= 7.5 mm), pinhole aperture (D= 3.2 mm), and a CCD (10 × 10 mm2). In the case of isotope heavy ions such as C-11, and O-15, the Cerenkov luminescence images showed a good correlation with the dose distribution (positional difference = 0.2 mm). However, when a C-12 beam was irradiated, the Cerenkov luminescence image showed no correlation with the dose distribution. In conclusion, the CLI has the potential to be used for the beam range monitoring of isotope heavy ions. In the future, experimental measurements will be performed to validate the GATE simulation results.

Keywords: Heavy ion therapy, Cerenkov luminescence imaging (CLI)
Poster panel: 403

Poster Number:

Proton CT: A potential solution for proton treatment planning in the presence of metallic implants (#1545)

C. Oancea1, 2, K. Shipulin1, G. Mytsin1, M. Gao3, M. Pankuch3, G. Coutrakon4, C. Ordonez5, R. Johnson5, V. Bashkirov6, R. Schulte6

1 Joint Institute for Nuclear Research, Dzhelepov Laboratory of Nuclear Problems, Dubna, Russian Federation
2 Horia Hulubei National Institute for R&D in Physics and Nuclear Engineering, CCR, Magurele, Romania
3 Northwestern Medicine Chicago Proton Center, Warrenville, Illinois, United States of America
4 Northen Illinois University, DeKalb, Illinois, United States of America
5 University of California at Santa Cruz, Physics Department, Santa Cruz, California, United States of America
6 Loma Linda University, Division of Bioengineering Sciences, Department of Basic Sciences, Loma Linda, California, United States of America


Imaging artifacts caused by metal objects, e.g. dental implants, can pose a serious problem for accurate proton treatment planning in head cancer patients. Such artifacts often require choosing beam directions that avoid the implants, but interference with artifacts cannot always be avoided and may cause relatively large range errors. In recent years, we have developed a proton CT (pCT) prototype and have tested it as a source of artifact-free planning CT images even in the presence of high-Z objects. The goal of this study was to compare the dose accuracy of pCT-based proton treatment planning with standard CT-based planning when two titanium implants were inserted into two adjacent teeth of an Alderson Radiation Therapy phantom. The phantom was scanned using both a standard CT and the pCT scanner prototype at the Northwestern Medicine Chicago Proton Center. These scans were imported into the clinical version of RayStation Treatment Planning System which uses a Monte Carlo dose calculation algorithm. Two simulated planning target volumes (PTV1 and PTV2) for each planning study were created. The PTV1 contained the implants at its center and the PTV2 was outside of the implants. A single-field-uniform-dose plan was calculated for both CT plans resulting in specific dose distributions for PTV1 and PTV2. A Gafchromic® EBT2 film was inserted between the phantom layers containing the Ti implants, and the phantom was exposed to each treatment plan. A 2D gamma index was calculated for each case (criterion 3mm, 3%). For PTV1 the pCT plan had a pass rate between 81.7% and 99.6%, whereas the standard CT plan had a pass rate between 42.2% and 83.5%. For PTV2, the pCT plan had a pass rate from 82% to 96.6%, whereas the standard CT had a pass rate, between 34.2% and 81.7%. The artifact-free image generated by the pCT method combined with an MC based TPS resulted in a single-field uniform dose distribution that was more accurate than the one planned with standard CT.

Keywords: Proton-based CT, dental implants, CT artifacts, Monte Carlo-based Treatment Planning System, proton dose distributions
Poster panel: 406

Poster Number:

Patient-specific voxel-based dosimetry using deep convolutional neural network  (#1640)

M. S. Lee1, D. Hwang2, 1, J. S. Lee2, 1

1 Seoul National University, Department of Nuclear Medicine, Seoul, Republic of Korea
2 Seoul National University, Department of Biomedical Sciences, Seoul, Republic of Korea


Personalized dosimetry with high accuracy is becoming more important due to the growing interests in personalized medicine. Dose calculation using direct Monte Carlo simulation is considered as the state-of-art voxel-based dosimetry technique but it requires extensive computational cost and time. In this study, we proposed to use the deep convolutional neural network (CNN) for the voxel dose prediction with extensively reduced computational time. In this study, the patch-based U-net architecture was used to learn dose rate maps from PET and CT images. Eight patient data sets, total 320,000 3D image patches, were used for the training and two patient data sets were used for testing. The predicted voxel dose rate maps from CNN were compared with the dose rate maps from direct Monte Carlo simulation using GATE which was considered as the ground truth. Moreover, dose rate maps generated from voxel S-value (VSV) kernel convolution method, which is most widely voxel-based dosimetry technique, was tested for the comparison. Finally, whole-body dosimetry study was conducted using CNN. The proposed CNN approach showed slightly denoised images but agreed well with the ground truth. The CNN approach showed reliable dose prediction results showing voxel dose errors of -0.18%±9.07%, while the VSV kernel approach showed voxel error -10.80%±16.50%. As a result of dosimetry results, CNN approach showed errors at the organ level under ±2%. The average organ dose percentage errors were -0.00093% and -7.56% for the CNN and VSV-based dosimetry results, respectively. The proposed method showed improvements compared to the conventional dosimetry approaches and showed comparable results with direct Monte Carlo simulation with significantly reduced dose calculation time.

Keywords: Radiation dosimetry, Deep convolutional neural network
Poster panel: 409

Poster Number:

A simulation study on background reduction using veto counters for imaging of therapeutic proton beams by measuring secondary electron bremsstrahlung (#1686)

M. Yamaguchi1, Y. Nagao1, N. Kawachi1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Takasaki Advanced Radiation Research Institute, Quantum Beam Science Research Directorate, Takasaki, Japan


A simulation study on reduction of a background (BG) component for an imaging method of therapeutic proton beams by measuring secondary electron bremsstrahlung (SEB) component was performed using the Monte Carlo method. Simulations were performed using PHITS version 2.96. In each simulation, a 139-MeV proton beam was injected to a water phantom consisted of a ceiling-opened container composed acrylic boards and water in the container. A parallel-hole-type X-ray camera was placed close to a water phantom. It consisted of an imaging detector, a radiation shield, and two veto counters. The imaging detector and the veto counters were made of cerium-doped gadolinium aluminum gallium garnet (GAGG). One of the veto counters, named “veto collimator” played a role of an active collimator and was placed between the water phantom and the imaging detector. The other veto counter, named “veto block” was placed so as to put the imaging detector between the veto collimator and the veto block. We simulated the images obtained by measuring the SEB and the BG component. We found that reduction of the BG component was possible by utilizing the veto counters. The ratio of the BG component to the sum of the SEB and BG components was able to be reduced from 73% to 20%. Even if using only the veto block, it was reduced to 45%.

Keywords: proton therapy, beam-trajectory imaging, secondary electron bremsstrahlung, Monte Carlo simulation, X-ray camera
Poster panel: 412

Poster Number:

Evaluation and Validation of a Sensitivity Model for a Three-layer LaBr3 Compton Telescope (#1730)

J. Roser1, E. Muñoz1, L. Barrientos1, J. Barrio1, J. Bernabéu1, M. Borja-Lloret1, A. Etxebeste1, L. Gabarda1, C. Lacasta1, G. Llosá1, A. Ros García1, J. F. Oliver1

1 IFIC (Instituto de Física Corpuscular), Experimental, Paterna (Valencia), Spain


A three-layer Compton telescope is being developed at IRIS group (Image Reconstruction, Instrumentation and Simulations for medical imaging applications) of IFIC-Valencia. It is based on continuous LaBr3 crystals coupled to SiPMs; consequently, a large number of projections can be obtained (high granularity). Precise estimations of both Sensitivity and System Matrix are needed for achieving an accurate reconstruction process. Given that Monte-Carlo calculations of the aforementioned quantities are computationally expensive due to the high granularity of the imaging device, an analytical model has been proposed. Whereas the validation of this model has been almost carried out for a two-layer Compton telescope, a similar effort must be made for the three-layer mode. Such validation must take into account the different topology of the events that form either signal or background in a three-layer Compton telescope, as well as the decrease of the efficiency. In this work we present the first preliminary tests on the validation of the proposed model, by comparing its performance and predictions with Monte-Carlo simulations. Images of simulated point-like monochromatic sources, as well as a preliminary background study, are also reported.

Keywords: Compton imaging, Background/signal ratio, Monte Carlo simulation, Tomographic Image Reconstruction
Poster panel: 415

Poster Number:

Research and Development of Real-time, 3D Dose Distribution Measuring Technique for Radiation Therapy Based on Novel Scintillation Gels (#1801)

H. Li1, M. Niu1, D. Li1, C. Wan2, X. X. Dai1, X. D. Zhang1

1 China Institute for Radiation Protection, Taiyuan, China
2 Lanzhou University, School of Nuclear Science and Technology, Lanzhou, China


A real-time and full 3D dosimetry is very important for the quality assurance (QA) of dynamic radiotherapy, such as the intensity-modulated radiation therapy (IMRT) and the volumetric-modulated arc therapy (VMAT). The purpose of this work is to research, develop, and validate a novel type of high resolution 3D dosimeter composed of soft tissue equivalent gels scintillator and an ultra-low noise and high sensitive light field camera. The scintillation gels were researched and developed, and the dose equivalent performance of the initially developed sample for γ-rays was measured experimentally. The results show that the differences between its mass attenuation coefficient of the gel and the recommended value of the soft tissue given by ICRU for γ-rays are about 10% for Eγ <150 keV and 0.1% for Eγ>150 keV, respectively. For neutron dose monitoring, the neutron-sensitive isotopes or elements need to be incorporated into the gel, such as 6Li and natural Ga, and this work is ongoing. The light yield of the developed gel scintillator can reach about 50% of the one of the standard liquid scintillators. There is still a room to further improve the light yield of such scintillator through fine tuning the contents of PPO and wavelength-shifting agent (bis-MSB). To read out the scintillation light from the gels, an ultra-low noise & high sensitive light field camera made of a micro-lens array, a relay lens, and an ultra-low noise scientific CMOS has been developed. Preliminary imaging test has been conducted by using a laser beam to illuminate the scintillation gels. 3D image reconstruction algorithm and dose integral calculation are now ongoing, and the corresponding results will be reported on the conference as well.

Keywords: Tissue Equivalent Scintillation Gel; Radiation Therapy; 3D Dose Distribution; Real-time Measurement; Dose Reconstruction
Poster panel: 418

Poster Number:

Monte Carlo simulation of the most likely path of protons through a transverse heterogeneity (#1940)

F. Khellaf1, N. Krah1, I. Rinaldi2, J. M. Létang1, S. Rit1

1 Univ Lyon, INSA-Lyon, Université Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR 5220, U1206, Lyon, France
2 University of Lyon, Université Claude Bernard Lyon 1, CNRS-IN2P3, Institut de Physique Nucléaire de Lyon, UMR 5822, Villeurbanne, France


Image reconstruction in proton computed tomography (pCT) requires the use of a most likely path (MLP) formalism taking into account the multiple Coulomb scattering to estimate proton trajectories. The MLP formalism uses a homogeneous medium assumption to compute proton paths. In this work, we test the accuracy of this formalism in heterogeneous media. More particularly, we study transverse heterogeneities causing unbalanced scattering. We have compared the paths computed using the MLP formalism to paths obtained from Monte Carlo simulations. Our results show transverse heterogeneities cause non Gaussian spatial and angular distributions, inducing an MLP that is different from the analytical MLP predicted by the formalism. This difference reaches 0.5 mm in the worst case, which is of the same order of magnitude as the formalism’s uncertainty, which is about 0.43 mm in water. We therefore deem the formalism accurate enough even in a transversely heterogeneous medium.

Keywords: most likely path, proton computed tomography, monte carlo simulation
Poster panel: 421

Poster Number:

Effect of Strong Load Variations on Gain and Timing of CeBr3 Scintillation Detectors Used for Range Monitoring in Proton Radiotherapy (#2169)

G. Pausch1, 2, C. Mueller2, J. Berthold3, W. Enghardt1, 2, M. Kuechler3, K. E. Roemer4, J. R. Stein5, A. Straessner3, A. Wagner4, T. Werner1, 2, A. Wolf5, T. J. Kögler1, 2

1 Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Institute of Radiooncology - OncoRay, Dresden, Saxony, Germany
2 OncoRay - National Center for Radiation Research in Oncology, Dresden, Saxony, Germany
3 Technische Universitaet Dresden, Institute for Nuclear and Particle Physics , Dresden, Saxony, Germany
4 Helmholtz-Zentrum Dresden - Rossendorf (HZDR), Institute of Radiation Physics, Dresden, Saxony, Germany
5 Target Systemelektronik GmbH & Co. KG, Wuppertal , North Rhine-Westphalia, Germany


CeBr3 scintillation detectors with light readout by photomultiplier tubes (PMT) have been used for prompt-gamma measurements in proton radiotherapy aiming at treatment verification by Prompt Gamma Timing (PGT) or Prompt Gamma Spectroscopy (PGS). Such treatments are usually structured in distinct beam spots grouped in several mono-energy layers, separated by breaks of few-seconds duration for beam energy switching. This causes a multitude of extreme load steps during delivery of a single treatment field. The paper presents preliminary results of two experiments exploring effects of such load steps on gain and timing of PGT detection units as developed for use in clinical treatments. Multiple units, consisting of scintillation detector (2” × 1…2” CeBr3 crystal coupled to Hamamatsu R13089 or R13089-100 photomultiplier) and a plugged-on high-throughput digital spectrometer (U100 by Target Systemelektronik) that also provides well stabilized dynode voltages for the PMT, were exposed to prompt gamma radiation produced by 225 MeV protons in a plexiglass (PMMA) layer, or to bremsstrahlung produced by 13 MeV electrons. Beam shots of 3-5 s duration and varied intensity provoked load steps from background up to the Mcps region and back, and allowed analyzing the immediate and the retarded response of PMT gain and timing. Indeed we observed a noticeable change of the PMT transit time with the detector load, indicating that space charge effects are involved. The scaling of gain turns with the mean anode current supports this hypothesis. As long as the mean anode current is in a ‘reasonable’ operating range, gain and timing drifts of given detectors are well correlated, at least in the stationary case. The observed load-induced timing shifts are as large as 100 ps and would seriously disturb PGT measurements in a treatment, but could eventually be corrected for by tracking the PMT gain. In treatments, this could be done by tracking the ubiquitous 511 keV annihilation peak.

Keywords: gamma spectroscopy, photomultiplier, load effect, proton therapy, range verification
Poster panel: 424

Poster Number:

Experimental studies of broadening in water of proton or carbon ion pencil beams for Hadron Therapy (#2350)

G. Gambarini1, 2, G. Barzon1, D. Bettega1, 2, L. Bettinelli1, G. Camoni1, M. Carrara3, A. Mirandola4, M. Ciocca4

1 UnUniversità di Milano, Department, Milano, Italy
2 INFN (National Institute of Nuclear Physics), Section of Milan, Milano, Italy
3 Fondazione IRCCS Istituto Nazionale Tumori, Radiation Oncology 2 Unit, Milano, Italy
4 Centro Nazionale di Adroterapia Oncologica (CNAO), Medical Physics Unit, Pavia, Italy


In this work, experimental results on parameters describing trend and broadening in water of proton and carbon ion pencil beams are reported. Measurements were performed by means of Gafchromic films, appropriately calibrated, to attain spatial information. The films were suitably placed in a water phantoms and exposed to proton or carbon ion beams at the Synchrotron of CNAO (Pavia). This experiment is aimed to get parameters for a software in MATLAB®, still in continuous development, focused to compensate for the quenching effects (due to radiation LET) in dose images acquired with films so to have the possibility of performing dose distribution controls in complex irradiations for patient therapy. In some irradiations, films were placed in planes normal to the pencil beam axis, at various depths. The same procedure was carried out for protons and carbon ions, with pencil beams of various energies. The transversal distributions of the measured absorbed doses were fitted by mean of a single Gaussian approximation for protons and a double Gaussian for carbon ions. The so obtained surfaces give correct information of the broadening of the beams, deduced by evaluating the section area at half height, and also allow to obtain an estimate of the quenching entity by applying a properly developed procedure of comparison between measured and calculated profiles of the absorbed dose. Some films were also placed, in the phantom, in a plane inclined by 5° relative to the beam axis and irradiated with pencil beams properly scanned to a homogeneous dose to a layer of 30 x 30 mm2. From the so obtained dose images, by means of properly developed software the in-depth images of the single pencil beams were achieved. The studies relating to carbon ions are more complex than those relating to protons, because the quenching effect is more consistent (also present at low depths) and moreover the transversal characteristics of the pencil beams can change from one time to another.

Keywords: Hadron Therapy, Carbon ions, Protons, Pencil beam broadening, Dose imaging
Poster panel: 427

Poster Number:

Effect of  time of flight in the monitoring of clinical proton beam treatments using the INSIDE PET detector (#2569)

N. Camarlinghi1, 2, G. Baroni5, 6, G. Battistoni7, N. Belcari1, 2, P. Cerello3, M. Ciocca6, M. D. De Rocha Rolo3, A. Del Guerra1, 2, M. Donetti6, V. Ferrero3, V. Gambetta1, 2, S. Giordanengo3, G. Giraudo3, A. C. Kraan1, 2, C. Luongo2, A. Mairani6, M. Morrocchi1, 2, S. Muraro2, F. Pennazio3, E. Fiorina3, 6, C. Peroni3, 4, A. Rivetti3, S. Rossi6, V. Rosso1, 2, G. Sportelli1, 2, S. Tampellini6, F. Valvo6, R. Wheadon3, M. G. Bisogni1, 2

1 Università di Pisa, Dipartimento di FIsica, Pisa, Italy
2 Istituto Nazionale di Fisica Nucleare, sez. Pisa, Pisa, Italy
3 Istituto Nazionale di Fisica Nucleare, sez. Torino, Torino, Italy
4 Università degli studi di Torino, Dipartimento di Fisica, Torino, Italy
5 Politecnico di Milano, Milano, Italy
6 Fondazione CNAO, Pavia, Italy
7 Istituto Nazionale di Fisica Nucleare, sez. Milano, Milano, Italy


PET is one of the most mature techniques to monitor ion beam treatments.  Recently, there has been a growing interest in ‘’in-beam’’ systems, i.e. systems that can be installed directly on the treatment gantry. However, this class of detectors, typically have incomplete angular coverage and therefore they produce images with severe artifacts. Several studies have shown that time of flight can mitigate the artifacts and improve the imaging capability of these systems. In this paper, we will show the result of the time of flight applied to the INSIDE PET. The experimental data reported in this paper were acquired when the INSIDE system was tested at CNAO, Pavia, the largest particle therapy clinical facility in Italy. In particular, we will show the impact of time of flight on the data of the first patient monitored with the INSIDE system. 

Keywords: PET, Hadron therapy, Time of flight
Poster panel: 430

Poster Number:

Feasibility of using Neutron Radiography in Proton Therapy (#2689)

S. Lee1, 2, M. Lu2, C. - W. Cheng3

1 Allegheny Health Network, Radiation Oncology, Pittsburgh, Pennsylvania, United States of America
2 Varex Imaging Corp, Santa Clara, California, United States of America
3 University Hospitals, Radiation Oncology, Cleveland, Ohio, United States of America


A neutron imaging flat panel detector (FPD) from PerkinElmer is investigated for its potential real time imaging application in proton therapy. Neutrons are produced by impinging a proton beam from a Mevion S250 unit on a brass plug (water equivalent thickness 30 cm) and a solid water phantom (9 cm thick). The FPD response characteristics to proton dose and dose rate and its imaging performance characteristics such as MTF and contrast resolution are characterized. The Leeds and the Las Vegas phantoms are used to determine the MTF and the contrast resolution of the FPD. The results are compared with those obtained with 6 MV x rays. The detector characteristics increase linearly with MU and MU/min. The MTF and the contrast resolution are comparable to those of the MV FPD from the Elekta Linac. Images of a key set obtained with neutron FPD produced in a solid water phantom illustrates the potential application of neutron imaging in proton therapy.

Poster panel: 433

Poster Number:

Research and development of neo beam ON-LINE PET system for dose-volume delivery guided proton therapy (#2935)

A. Nishio1, T. Nishio1, T. Okamoto2, S. Kabuki3, H. Yamashita4, T. Mizowaki5, M. Tsuneda1, T. Masuda1, K. Karasawa1

1 Tokyo Women’s Medical University, Tokyo, Japan
2 Hamamatsu Photonics K. K., Shizuoka, Japan
3 Tokai University, Kanagawa, Japan
4 Shizuoka Cancer Center, Shizuoka, Japan
5 Kyoto University, Kyoto, Japan


Proton therapy is a form of radiotherapy that enables the concentration of a dose onto a tumor by the modulation of Bragg peak. Therefore, it is very important to evaluate the proton-irradiated volume accurately. The dose-volume delivery guided proton therapy (DGPT) can be confirmed by detection of pair-annihilation gamma rays from positron-emitting nuclei generated in the patient body by the target nuclear fragment reaction of the incident protons on target nuclei. In this study, we designed and developed neo beam ON-LINE positron emission tomography (PET) system (nBOLPs) with imaging technique in activity of generated positron-emitting nuclei for achievement of the DGPT. The nBOLPs is constructed with 4 planar type detector heads that the 36 detector units mounted on each detector head and each unit was composed of 10 x 10 arrays of GSO (Gd2SiO5(Ce)) crystals with a crystal size of 2 mm x 2 mm x 25 mm. Therefore, 4 planar type detector heads have 14,400 crystals. Furthermore, special software forimage reconstruction algorithm, image viewer tool and physical analysis tool was developed for activity imaging. Performance of the nBOLPs was verified with use of activity source ofpair-annihilation gamma raysand clinical proton beam. Verification results were imaging volume of about 15 cm x 15 cm x 15 cm, detection-position resolution better than 3 mm, and detection rate better than 3 kcps. And the experiment of proton beam irradiation to head & neck and pediatric phantoms was carried out for evaluation in clinical situation of proton therapy. The developed nBOLPs has great potential for high precision treatment with proton irradiated volume imaging and DGPT in the future.

Keywords: Proton therapy, nBOLPs, target nuclear fragment reaction, activity imagimg, DGPT