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: Choi, Yong, (Sogang University, Molecular Imaging Research & Education (MiRe) Laboratory, Department of Electronic Engineering, Seoul, South Korea); Lois, Cristina, (Massachusetts Institute of Technology, US)
Shortcut: M-14
Date: Friday, 16 November, 2018, 10:20 AM
Room: Park Side Ballroom
Session type: MIC Session


Click on an contribution to preview the abstract content.

Poster panel: 3

Poster Number:

Performance Comparison of Depth-Encoding Detectors Based on Dual-Ended Readout and Different SiPMs for Brain PET Applications (#1275)

J. Du1, X. Bai1, S. R. Cherry1

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


The performance of four dual-ended readout detector modules based on SiPM arrays from Hamamatsu, KETEK and SensL were evaluated and compared for brain PET applications.  The SiPM arrays are all 8 x 8 array of SiPMs. The SiPM arrays of Hamamatsu have a surface area of 25.6 mm x 25.6 mm and a pitch of 3.2 mm, and the SiPM arrays of SensL and KETEK have a surface area of 26.88 mm x 26.88 mm and a pitch of 3.36 mm. In each detector module, two identical SiPM arrays were coupled to both ends of a 24 x 24 array of LYSO elements with a crystal pitch of 1.0 mm and a length of 20 mm. Toray was used as the reflector to separate the LYSO elements. Custom front-end electronics were designed to reduce the 128 SiPM signals of each dual-ended detector module to nine signals (eight for position information and one for timing information). The performance of each detector module in terms of flood histogram, energy resolution, DOI resolution and timing resolution were obtained at different bias voltages and four different temperatures (0 ̊C, 10 ̊C, 20 ̊C and 30 ̊C).

Keywords: PET, DOI
Poster panel: 6

Poster Number:

Performance Comparison of Two PET Detectors Based on SiPM Arrays with Same Pitch Size and Different Active Areas (#1325)

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

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


The performance of two dual-ended readout PET detectors for a high-resolution and high-sensitivity MRI-compatible brain PET insert was evaluated and compared. One detector is based on 8 × 8 arrays of SensL MicroFJ-30035 SiPMs with 3.07 × 3.07 mm2 active area per pixel, and the other one is based on 8 × 8 arrays of SensL MicroFJ-40035 SiPMs with 3.93×3.93 mm2 active area per pixel. The pitch size of both SiPM arrays is 4.2 mm, leaving different dead spaces between SiPM pixels. The two detectors are fabricated by coupling two identical SiPM arrays to both ends of a 20×20 LYSO array with a crystal size of 1.42 mm ×1.42 mm × 20 mm. The pitch size of the LYSO array is 1.5 mm and BaSO4 with a thickness of 80 um is used as reflector. Performance in terms of flood histogram, energy resolution, signal-to-noise ratio, depth-of-interaction resolution and timing resolution were measured at different bias voltages from 27.0 V to 30.0 V with an interval of 0.5 V and a fixed temperature of 22.8 °C. The results show that all crystals can be resolved from the flood histograms of the two detectors and the detector using MicroFJ-40035 SiPMs provides a better flood histogram quality. The best energy resolution, signal-to-noise ratio, DOI resolutions and timing resolution are 12.6%, 101, 2.07 mm and 986 ps for detector using MicroFJ-30035 SiPMs, and 10.9%, 117, 2.21 mm and 822 ps for detector using MicroFJ-40035 SiPMs. Both detector modules provide promising performance and detector using MicroFJ-40035 SiPMs provides a better overall performance.

Keywords: PET detector; SiPM; depth of interaction; brain imaging
Poster panel: 9

Poster Number:

Design of a Compton Scatterer Detector with Lateral Position-Encoding Capability (#1382)

Z. Wang1, 2, M. - L. Jan2, 3, S. Y. Wang2, 3, H. Y. Chen4, M. - W. Lee2, 3, K. S. Chuang1, J. - H. Hong3

1 National Tsing Hua University, Department of Biomedical Engineering and Environmental Sciences, Hsinchu, Taiwan
2 Chang Gung University and Chang Gung Memorial Hospital, Institute for Radiological Research, Medical Physics Research Center, Taoyuan, Taiwan
3 Chang Gung Memorial Hospital, Department of Radiation Oncology, Taoyuan, Taiwan
4 Chang Gung University, Department of Medical Imaging and Radiological Sciences, Taoyuan, Taiwan


A prompt-gamma Compton camera (PGCC) with scintillator-based side-on scatterers is being developed at our laboratory to provide improvements in sensitivity, and spatial resolution for the proton range verification. To ensure the lateral field of view (FOV) of the PGCC, the side-on scatterer is composed of lateral-oriented long crystals with two ends coupled to the photosensitive areas of SiPMs. Since the spatial resolution along the axis of crystal length is usually the worse, we therefore have proposed a light sharing and neural network (LS-NN) technique to improve the lateral resolution of PGCC using long crystals. To evaluate the performance of LS-NN designs, two groups of 3 x 3 crystal arrays of unpolished 1.8 × 1.8 × 50 mm3 and 1.8 × 1.8 × 100 mm3 LYSO were examined. Each group consists of three crystal arrays which were partially covered with three types of inter-crystal reflector, namely arrow-shape type (AS), isosceles triangular-shape type (ITS), and opposite triangular-shape type (OTS). A learning-based neural network algorithm was employed to characterize the light distribution of the three partial-covered reflector types and their relationship with the position of interaction (POI). In this study, Monte Carlo simulations were performed for the evaluation of the proposed LS-NN designs. The results show the best average POI resolution is 0.73 mm FWHM, 2.03 mm FWHM of 50 mm, 100 mm long crystals, respectively. Both of them are with the arrow-shape type reflector. The LS-NN method provides not only the better POI resolution but also the excellent linearity of POI encoding capability. In other words, the LS-NN can overcome the crystal-end POI unresolved problem. Thus the side-on scatterer with the LS-NN design could enlarge the POI resolvable range. The AS-typed scatterer configuration with the proposed LS-NN position encoding shows a great promise to the development of 3D PGCC for proton range verification.

Keywords: Compton camera, machine learning, prompt gamma, position of interaction, proton therapy
Poster panel: 12

Poster Number:

Depth encoding PET detectors using single layer crystal with different reflector arrangements along depths (#1514)

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

1 Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China


Depth encoding detectors are preferred for PET scanners with small ring diameter and long axial field of view. In this work, novel depth encoding detectors using different inter-crystal reflector arrangements along the depths of single layer crystal arrays are proposed. As compared to the previous four-layer detector with different reflector arrangements, the detectors proposed in this work are expected to improve the energy resolution and timing resolution and easy to assemble in a PET scanner. 2×2 LYSO arrays using four and two reflector arrangements along the crystal depth are fabricated. The LYSO arrays are single-ended read out by SiPM arrays. The flood histograms and energy resolution of the detector were measured in both singles and coincidence modes. The detector with four reflector arrangements nearly provides four-layer depth encoding capability. Except the top two quarters which cannot be clearly resolved, the bottom three quarters of the detector can be clear resolved from the flood histogram. The detector with two reflector arrangements not only provides two-layer depth encoding capability, but also provides depth encoding capability within each half. The DOI resolution at the top of the detectors is worse than that at the bottom of the detector for both detectors. The DOI resolution of the detector with four reflector arrangements is better than that of the detector with two reflector arrangements. The average energy resolutions of the two detectors are 12.2% and 18.0%. In the next, large LYSO arrays will be made to further prove the feasibility of the detectors.

Keywords: depth-encoding PET detectors; single-layer crystal array; inter-crystal reflector; single-ended readout; DOI resolution
Poster panel: 15

Poster Number:

Optical imaging of produced light in water during irradiation of gamma photons lower energy than the Cerenkov-light threshold (#1558)

S. Yamamoto1, K. Kato1, S. Abe2

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


It is commonly believed that gamma photon irradiation to water does not produce light in water if the gamma photon energy is lower than the Cerenkov-light threshold and imaging of gamma photon is impossible because the produced electrons from gamma photons do not emit light. However, we found that the light is produced in water and the imaging of produced light in water was possible with gamma photon irradiation even at lower energy than the Cerenkov-light threshold. We irradiated gamma photons with lower energy than the Cerenkov-light threshold (260 keV for electrons in water) from 99mTc (gamma photon energy of 140.5 keV) to water phantom, optical imaging was carried out with a charge-coupled device (CCD) camera. Also we conducted imaging of the gamma photons irradiation to acrylic and plastic scintillator blocks. The optical images of water were observed from 3600 s exposure time, and the intensity of the luminescence was linearly increased with time although the intensity was ~0.001 % of that with the plastic scintillator block. The optical imaging of water during irradiation of gamma photons lower energy than the Cerenkov-light threshold was possible and it could be a new method for distribution measurements of gamma photons in water.

Keywords: Tc-99m, optical imaging
Poster panel: 18

Poster Number:

Visualization of Ra-223 chloride using an omnidirectional gamma-ray imaging Compton camera for radioactive environmental monitoring (#1585)

H. Muraishi1, R. Kondo1, T. Watanabe1, 2, R. Enomoto3, H. Katagiri4, M. Kagaya5, D. Kano6, T. Takeda1, S. Noda7, K. Kikuchi7, H. Hara1, T. Maegaki7, T. Uchida8, M. Tanaka8

1 Kitasato University, School of Allied Health Sciences, Sagamihara, Japan
2 Tokyo Metropolitan University, Department of Radiological Sciences, Tokyo, Japan
3 University of Tokyo, Institute for Cosmic Ray Research, Chiba, Japan
4 Ibaraki University, College of Science, Mito, Japan
5 National Institute of Technology, Sendai College, Department of General Engineering, Sendai, Japan
6 National Cancer Center Hospital East, Chiba, Japan
7 Kitasato University Hospital, Sagamihara, Japan
8 High Energy Accelerator Research Organization, Ibaraki, Japan


We investigated the possibility of visualizing low-level radioactive contamination by Ra-223 chloride injection in medical facilities using a highly sensitive gamma-ray omnidirectional imaging Compton camera. The detector consisted of CsI(Tl) scintillator cubes of 3.5 cm in size, which acted as scatterers or absorbers. The measurements were done using the detector in the nuclear medicine facility, where the Ra-223 sources (used vial, syringe, etc.) with the radioactivities of less than 1 MBq were placed inside the preparation room. We confirmed four groups of gamma-ray full-absorption peaks between 270 keV and 830 keV as expected in the energy spectrum made by the sum of energy deposits in two counter coincidence events. Furthermore, we successfully detected the direction of the Ra-223 source in the reconstructed gamma-ray image. Our proposed technique would be used as a low-level imaging dose monitor to visualize radioactive contamination by Ra-223 in medical facilities.

Keywords: Gamma-ray detector, Targeted alpha radiotherapy
Poster panel: 21

Poster Number:

PETALO: Time-of-Flight PET with liquid xenon (#1595)

P. Ferrario1, V. Herrero2

1 Donostia International Physics Center, Donostia-San Sebastián, Spain
2 Universidad Politécnica de Valencia-CSIC, I3M, Valencia, Spain

On behalf of the PETALO Collaboration.


Liquid xenon has several attractive features, which make it suitable for applications to nuclear medicine, such as high scintillation yield and fast scintillation decay time. Moreover, being a continuous medium with a uniform response, liquid xenon allows one to avoid most of the geometrical distortions of conventional detectors based on scintillating crystals. In this paper, we describe how these properties have motivated the development of a novel concept for positron emission tomography scanners with Time-Of-Flight measurement, which combines a liquid xenon scintillating volume and silicon photomultipliers for the readout. A first Monte Carlo investigation has pointed out that this technology would provide an excellent intrinsic time resolution, down to 70 ps, which makes it  possible to measure the Time-Of-Flight with high efficiency. Also, the transparency of liquid xenon to UV and blue wavelengths opens the possibility of exploiting both scintillation and Cherenkov light for a high-sensitivity positron emission tomography scanner with Time-Of-Flight capabilities. Monte Carlo simulations point to a time resolution of 30-50 ps obtained using Cherenkov light. A first prototype is being built to demonstrate the high energy, spatial and time resolution of this concept, using a ring of 20 cm of internal diameter and a depth of 3 cm instrumented with VUV--sensitive silicon photomultipliers.


Poster panel: 24

Poster Number:

Pilot studies towards positronium imaging with the total-body PET scanners (#1831)

P. Moskal1

1 Jagiellonian University, Institute of Physics, Cracow, Poland



The purpose of the reported research is the elaboration of the new imaging method based on the in-vivo measurement of properties of positronium produced inside patient during positron emission tomography, and determination of correlations between properties of positronium inside the cancer tissues and  histopathological characteristics of cancers.

During PET diagnosis positronium may be trapped inside free volumes between molecules of the examined patient. Currently, in the PET technique, the phenomenon of positronium production is neither recorded nor used for imaging. Yet in more than 40% cases, the electron-positron annihilation proceeds in the tissue via creation of positronium. We will present (i) results of the feasibility studies of the positronium mean-lifetime image reconstruction with the total-body PET scanner from plastic scintillators, as well as (ii) results of pilot studies of the mean lifetime of positronium in the healthy and tumorous tissues operated from the patients.  Performed experiments show that properties of positronium atoms in uterine tissues operated from human patients reveals meaningful differences between healthy and tumorous tissues (larger than 50 ps). Moreover, the performed simulations show that in the future with the total-body PET and improved time resolution it shall be feasible to  reconstruct images of positronium mean lifetime  with the precision enabling to observe differences in lifetime of positronium between the normal and tumorous tissues.   

Keywords: positronium imaging, total-body PET
Poster panel: 27

Poster Number:

Timing Resolution Improvements from Cooling in BGO PET Detection with Analog SiPMs (#1944)

A. K. McVea1, C. - M. Kao1, C. - T. Chen1

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


Using Bismuth Germanate (BGO) scintillators coupled to analog Silicon Photomultipliers (SiPM) we are developing methods to improve the coincidence time resolution (CTR) of PET detectors with BGO in order to reach a CTR that can be used for time of flight (TOF) PET detection. TOF measurements have already been acquired using LYSO crystals, but the low cost of BGO along with its high density make BGO an attractive option to reduce the overall cost of TOF PET scanners. Using an analog SiPM with a pre-amplifier at a higher voltage improves energy and timing resolution for BGO coincidence detection, but also introduces a much higher baseline noise level, which can obscure the start of the waveform. We have found this noise level can be significantly reduced by cooling the detectors to values near -25oC, allowing us to perform leading edge discrimination (LED) timing at much lower voltages. On average cooling the detectors to this point improves their timing resolution by around 300ps, and as much as 600ps. Using this setup our values are approaching the values necessary for TOF PET detection. In this paper we will discuss the experimental setup in further depth and share promising initial timing and energy resolution results from our findings.

Keywords: PET, BGO, TOF
Poster panel: 30

Poster Number:

Development of Depth-of-Interaction PET Detectors using Novel Ceramic Garnet Scintillator (GLuGAG:Ce) (#1978)

S. I. Kwon1, P. Peng1, Y. Wang2, J. Glodo2, U. Shirwadkar2, K. S. Shah2, S. R. Cherry1

1 University of California, Davis, Genome and Biomedical Sciences, Davis, California, United States of America
2 Radiation Monitoring Devices, Inc.,, Watertown, Massachusetts, United States of America


To achieve both uniform spatial resolution and high sensitivity, depth-of-interaction (DOI) capability is required, especially for dedicated breast, brain, or long axial field of view positron emission tomography (PET) scanners. Dual-ended readout is one of the more popular depth encoding methods. In this method, the ratio of signals from opposing ends is used to calculate depth. To improve DOI resolution, rough surfaces of scintillators and less efficient reflector material have been favored. However, in this case, energy and timing resolution are usually degraded. We have developed and investigated a novel transparent ceramic garnet scintillator, (GdxLu1-x)3(GayAl1-y)5O12:Ce abbreviated as GLuGAG. It has a higher light yield than LSO, high stopping power, and good timing properties. Since GLuGAG has a cubic crystal structure, it is fabricated using ceramic techniques, resulting in a major reduction in cost compared with LSO. Although the light yield of GLuGAG is higher than that of LSO, GLuGAG has some self-absorption, which can be exploited to positive effect using the dual-ended readout method to determine DOI. The self-absorption helps change the ratio of the signals detected at opposing ends of the GLuGAG material in a depth-dependent fashion. In this study, we present the first GLuGAG pixel arrays for DOI PET detectors and preliminary results of the arrays coupled to SiPM arrays (J-series array, Sensl). Each pixel was polished, and enhanced specular reflector film (ESR) and polyester film (Toray) were used for reflector materials. A dedicated 64 channel ASIC (TOFPET2) and front-end board were used to digitize each SiPM signal. GLuGAG ceramics showed good performance for DOI PET detectors; average DOI resolution was less than 4 mm with Toray reflector. DOI resolution was much improved through the individual readout of SiPM outputs and the selective use of SiPM outputs to calculate DOI. Different SiPM arrays will also be evaluated, and results will be compared.

Keywords: PET, Ceramic, Depth-of-interaction
Poster panel: 33

Poster Number:

Primarily evaluation of a TOF-PET detector based on Petiroc2A ASIC (#2050)

Z. Lu1, 2, X. Huang1, W. Zhou1, Y. Wang1, H. Tang1, Z. Zhang1, D. Li1, L. Li1, L. Shuai1, Y. Zhang1, L. Wei1, C. Ma1

1 Beijing Engineering Research Center of Radiographic Techniques and Equipment, Institute of High Energy Physics, Chinese Academy of Sciences, Beijing, China
2 School of Nuclear Science and Technology,University of Chinese Academy of Sciences, Beijing, China


Time of flight (TOF) information could be very helpful in improving the PET reconstructed image quality. In this work, detector module based on SiPM and readout electronics based on Petiroc2A Application Specific Integration Circuits (ASIC) were developed. The preliminary tests were carried out in order to validate its use for TOF-PET systems. Two detector modules in the opposite direction are used in this system and coupled with each other. Each detector module consist of a 4×4 pixelated LYSO crystal array (pixel size: 3×3×20mm3) on the top of a SensL’s J-30035 4×4 SiPM array. A full digital measurement board based on Petiroc2A ASIC was developed to readout and process the signals from the pair detector modules with a coincidence mode. Xilinx Virtex-5 FPGA and gigabit network were used to configure and control the Petiroc2A chip. Online data analysis and configuration was achieved by the use of dedicated LabWindows software. The preliminary results showed that this detector system has the ability to achieve 417ps coincidence time resolution with two corresponding crystals and 1.6ns coincidence time resolution with two corresponding crystal arrays. After the energy linear calibration, the energy resolution at 511keV channel is about 9.7%. It suggests that the future of this detector is promising for TOF-PET applications.

Keywords: Time of flight; Detector; SiPM; Petiroc2A ASIC
Poster panel: 36

Poster Number:

A new in-beam proton therapy monitoring system based on digital MVT readout (#2155)

N. D'Ascenzo1, M. Gao1, H. - H. Chen2, 3, F. - H. Chen2, J. - H. Hong3, 5, I. - T. Hsiao2, T. - C. Yen4, W. Wang8, D. Xi6, B. Zhang7, Q. Xie1

1 Huazhong University of Science and Technology, the School of life science and technology, Wuhan, China
2 Chang Gung University, the Department of Medical Imaging and Radiological Science, Taoyuan, China
3 Chang Gung Memorial Hospital at Linkou (CGMH at Linkou), the Department of Radiation Oncology, Taoyuan, China
4 Chang Gung Memorial Hospital/Chang Gung University (CGMH/CGU), the Molecular Imaging Center and Nuclear Medicine, Taoyuan, China
5 Chang Gung Memorial Hospital/Chang Gung University (CGMH/CGU), the Institute for Radiological Research, Taoyuan, China
6 RayMeasure Medical Technology Co., Ltd., Suzhou, China
7 RaySolution Digital Medical Imaging Co., Ltd., Ezhou, China
8 Ministry of Industry and Information Technology, the Key Laboratory of Biomedical Engineering and Translational Medicine, Beijing, China


We realized a prototype of a monitoring system for proton therapy based on the technique of positron emission tomography. For the first time to our knowledge, we used the Plug&Imaging (P&I)Multi Voltage Threshold (MVT) technology in this application. The sensing system includes LYSO/silicon photomultiplier (SiPM) detection elements, fast digital multi voltage threshold (MVT) readout electronics and dedicated image reconstruction algorithms. In this paper we show that the P&I sensor system can be operated in the in-beam beam-on experimental conditions of the proton therapy room. The saturation count rate obtained in this study during the beam-on time is 5.2~Mcps and overwhelms  the current reported limits of other studies significantly. We demonstrated that the prototype is able to be operated during beam irradiation and provides a spatial reconstruction of the proton-induced positron emitters  during both the irradiation (beam-on) high prompt photon background state and after the irradiation in the absence of prompt-photons . Finally, the time stamping obtained with the MVT method applied to the  LYSO/SiPM detection system allows to reconstruct both the decay and the relative abundance of the positron-emitters  produced in the phantom after proton irradiation with  relative accuracy of 0.5% and 5% respectively. Our results are in agreement with previous studies performed using SiPM-based PET monitoring systems.

Keywords: Silicon sensors, digital readout, multi voltage threshold, positron emission tomography, proton therapy
Poster panel: 39

Poster Number:

Reduction of readout channels in PET detectors preserving performance capabilities (#2250)

A. González-Montoro1, A. Perez2, 1, L. Hernandez1, E. Lamprou1, J. Barrio1, G. Cañizares1, M. Freire3, 1, S. Sanchez1, F. Sánchez1, L. F. Vidal1, V. Ilisie1, J. M. Benlloch1, A. J. González1

1 Universitat Politècnica de València, Instituto de Instrumentación para Imagen Molecular (I3M), Valencia, Spain
2 Universidad Valencia, Escola Tecnica Superior Enginieria , Burjassot, Spain
3 Universidad Valencia, Burjassot, Spain


In this work we show a method to reduce the number of signals to be processed from a detector block suited for gamma ray imaging. Several types of photosensors are currently in use depending on the application, most known are PSPMT or SiPM arrays. There have been investigations towards reducing the number of signals. Anger logic is one of widest used approach, since it reduces the number of photosensor signals to only 4. Other approaches result into reducing the number of signals to the number of rows and columns in the photosensor array.

We show here a two-steps reduction scheme. On one step, we merge signals at the center and laterals of the detector block where less sampling is typically required. The current approach allows reducing the 144 signals (12x12 SiPM array) to only 64. On a second reduction step, a further reduction is applied by projecting the already reduced signal into X and Y projections. In the particular case of an array of 12x12 photosensors, only 8+8 signals are finally obtained. We have tested this read-out with a thick 50x50x15 mm monolithic crystals. We have compared the results with standard row and column reduction schemes without observing significant deterioration on spatial resolution, energy or depth of interaction capabilities. Moreover, we have also compared the results with an alternative method on which, after using the traditional row and column projection, we have also reduce those by merging the central ones, also returning only 8+8 signals. We have also made tests with LYSO crystal arrays 1.6 mm, allowing one to resolve all 32x32 pixels.

Currently, using only the first step of the reduction readout (144 to 64), we have connected to the PETsys ASIC in order to evaluate the timing capabilities when merging signals. We expect very low timing degradation and, therefore, have the possibility to provide a reduction signals scheme keeping good performance with regards to spatial, energy, depth of interaction but also timing.

Keywords: PET readout, monolithic crystals, analog readout, reduction readout, SiPM arrays readout
Poster panel: 42

Poster Number:

PET detector combining monolithic crystals and ASIC readout (#2346)

E. Lamprou1, A. Aguilar1, G. Cañizares1, A. González-Montoro1, V. Ilisie1, J. Barrio1, F. Sánchez1, A. J. González1, J. M. Benlloch1

1 Universitat Politècnica de València, Instituto de Instrumentación para Imagen Molecular (I3M), Valencia, Spain


The aim of this work is to show the potential capabilities of monolithic crystals, coupled to large SiPM arrays for PET scanners enabling TOF capabilities. Monolithic blocks allow one to decode the 3D photon impact position. This can be of high interest in clinical PET design geometries where a typical ring configuration is not used, as well as in Total Body PET systems leading to a significant improvement of system resolution.

It has been demonstrated that the best approach in order to extrapolate accurate TOF information, is to read and process each photosensor element independently. We have evaluated the ASIC chip, named TOFPET2, that can handle up to 64 channels per block. Spatial and energy resolution tests for both crystal arrays and monolithic blocks were carried out and are presented in this work. Initial characterization of the ASIC CTR capabilities was carried out using small crystal pixels and SiPM photosensors. A time resolution near 200 ps FWHM was measured along with good energy resolution, revealing the limits in terms of CTR.

Providing timing resolution when using detectors based on monolithic and analog SiPMs is however challenging. The wide distribution of scintillation light causes a limited SNR, which makes the system sensible to false triggering and to time walk errors. In this direction, we present a calibration method, designed to correct all recorded timestamps and also to compensate variations in time-paths among channels. Thereafter, a CTR improvement of 35% is observed. Moreover, we show a novel approach which describes the use of weighted averaging methods to assign the timestamp to each gamma impact. This approach results in a further improvement of the CTR in the range of 100 ps FWHM, reaching a time resolution under 800 ps FWHM using large 50 x 50 x 15 mmscintillators coupled to 8 x 8 SiPM (6 x 6 mm2) arrays. These pilot studies show detector capabilities regarding TOF information when using monolithic scintillators.

Keywords: ASIC readout, SiPM readout, Time-of-flight
Poster panel: 45

Poster Number:

Material decomposition with a fast photon counting linear array detector (#2419)

A. Brambilla1, C. Paulus1, F. Mathy1, V. MOULIN1, L. Verger1

1 CEA, CEA LETI, MINATEC Campus, Grenoble, France


The emergence of CdTe Photon Counting Detectors (PCD) with energy discrimination capabilities offer new perspectives in X-ray imaging. They provide energy dependent information which can exploited for material identification. In the medical field, it is used for the quantification of tissue composition and contrast agent concentration.

In this study we show images obtained with a ME100 detector provided by MULTIX. The ME100 is a linear array pixel detector based on a of 3 mm thick CdTe sensor with 800 µm pixels. It provides spectral images made of high resolution spectra on 64 energy bins for each pixel. The spectral images processed by a basis material decomposition algorithm. Different examples of applications show the interest of spectral imaging to separate soft tissue and bone structures and to provide quantitative information on the composition of these tissues, such as bone density. In the presence of a contrast agent, the ME100 detector provides the 3 equivalent images with a single acquisition. The linear behaviour of the equivalent lengths solves the beam hardening artefacts during tomographic reconstruction.

Keywords: Photon Counting Detector, Spectral imaging, material decompostion
Poster panel: 48

Poster Number:

Improvement of the acquisition rate of a novel Digital PET helmet (#2452)

E. Antonecchia1, 2, N. D'Ascenzo1, 2, M. Gao1, P. Gnudi1, 2, Q. Xie1, 2

1 HUST - Huazhong University of Science and Technology, School of Life Science and Technology, Wuhan, China
2 NEUROMED IRCSS, Pozzilli (Is), Italy


Data acquisition systems and readout electronics play a fundamental role in the optimization of complex nuclear radiation systems. 

Our group has previously proposed a fast readout system for scintillator/SiPM radiation detection modules. It is based on the digital multi voltage threshold (MVT) sampling method that takes samples of a pulse with respect to a set of four reference voltages. The total energy and the timing of the pulse can be obtained by the digital signal processing techniques. The combination of novel LYSO/SiPM sensor technology, MVT dedicated digital electronics and image reconstruction constitutes the Plug\&Imaging (P\&I) sensor.
In this paper we present for the first time to our knowledge an improved readout strategy for the digital Plug\&Imaging PET detector system based on Multi Voltage Threshold method. We study different readout through Ethernet configurations and we obtain a maximal count rate ranging between approximately 5 Mcps and 8 Mcps, which overwhelms  the current reported limits of other studies significantly. We apply this scheme to an idea of digital PET Helmet, we test the electronics chain on a dedicated prototype and finally we support our results with a detailed GEANT4 simulation, which includes the features of the digitization chain. 

Keywords: Plug\&Imaging detector system, digital PET Helmet, GEANT4 simulation
Poster panel: 51

Poster Number:

Development of a high-resolution detector for a head-and-neck cancer dedicated PET scanner (#2554)

M. Li1, S. Abbaszadeh1

1 University of Illinois at Urbana-Champaign, Department of Nuclear, Plasma, and Radiological Engineering, Urbana, Illinois, United States of America


Whole body (WB) positron emission tomography/computed tomography (PET/CT) has become a standard clinical imaging modality in patients with head and neck cancers (HNC). PET/CT is used to establish how far the tumor has invaded locally and to guide the decision to resect a tumor rather than irradiate and give chemotherapy. Therefore, a high spatial resolution is especially important in the neck because layers of tissues are so thin. Current WB PET spatial resolution is ~ 4-6 mm. We design a two-panel dedicated HNC PET scanner and the proposed design shows a good performance in terms of noise equivalent rate, photon coincidence sensitivity, spatial resolution and lesion visualization. To implement the superior performance of the proposed system, we are developing a compact-geometry, high spatial resolution, high depth-of-interaction (DOI) resolution detector module with multiple scattering events recovery capability. Ce:LYSO is attached to SiPM (Hamamatsu 3050AE-08) and we use commercially available PETsys SiPM Readout System as the readout electronics. The LYSO pixel size is 1*1*20 mm3, while SiPM channel size is 3*3 mm3. One detector module is 6*6 mm2, which consists of 2*2 SiPM channels and 6*6 LYSO pixels. The roughness of the four side surfaces of LYSO pixels are ground to 10 μm and the two end surfaces of LYSO pixels are polished. The reflector material is Toray E60. The depth-of-interaction capability is implemented by dual-ended readout. To investigate the noise level of the SiPM and the readout electronics, we characterize a single 3*3*20 mmLYSO with one SiPM channel and a 100-μCi Na-22 point source is used. The single-channel energy resolution is measured as 7.9% FWHM. We will report the flood histogram, energy resolution, coincidence time resolution and DOI resolution of the detector module.


Keywords: high-resolution, scattering recovery, depth of interaction, head and neck cancer
Poster panel: 54

Poster Number:

MoTI, a mobile gamma camera for therapeutic dose control during targeted radiotherapy (#2673)

C. Trigila1, C. Esnault1, L. Pinot1, A. Desbrée3, T. Beaumont3, D. Broggio3, M. - A. Verdier2, Y. Charon2, M. - A. Duval1, D. Gray4, J. - P. Roussin4, L. Ménard2

1 Univ. Paris Sud et Paris Diderot, IMNC CNRS/IN2P3, Orsay, France
2 Univ. Paris-Diderot, IMNC CNRS/IN2P3, Orsay, France
3 Laboratory of Internal Dose Assessment, IRSN, Fontenay-aux-roses, France
4 AG-Medical, Saint-Aubin, France


Targeted radionuclide therapy is the most used treatment modality against malign and benign diseases of thyroid. The large heterogeneity of therapeutic doses in patients and the range of effects observed state that an individualized dosimetry is essential for optimizing this therapy. The goal of the project is to strengthen the control of the doses delivered to thyroid during treatment of benign and malign diseases, providing a novel mobile gamma imaging device specifically dedicated to measurements of the bio-distribution and kinetics of the radio-tracer at the patients’s bedside.

We report the optimization of the detection head of the camera, made by both experiments and Monte Carlo simulations, and the preliminary experimental results obtained with the first fully operational 5×5 cm2 FoV camera prototype. It consists of a 3D printed parallel-hole high-energy tungsten collimator, coupled to a 6 mm thick continuous CeBr3 scintillator, readout by an array of Silicon Photomultiplier detectors. The camera exhibits an intrinsic spatial resolution of 0.8 mm FWHM at 356 keV with very low distortion and an energy resolution of 8%. The optimization of the collimator design, in order to enhance small nodules detectability by reducing scatter and septal penetration, leads to the choice of a 5.5 cm thick collimator with a spatial resolution of 2 mm and an efficiency of 1.24×10−5 for a 5 cm source distance. Preliminary imaging with thyroid phantoms filled with 131-I shows the huge improvement of image quality compared to a standard high-energy gamma-camera. Detailed description of the MoTi camera optimization and the complete evaluation of its global performances (uniformity, resolution and efficiency as a function of the source distance, quantification with thyroid phantoms) will be presented.

Keywords: targeted radionuclide therapy, 131I, gamma imaging
Poster panel: 57

Poster Number:

Optimized TOF-PET Detector Using Scintillation Crystal Array for Brain Imaging (#2792)

H. Leem1, Y. Choi1, J. Jung1, K. Park1, Y. Kim1

1 Sogang University, Electronic Engineering, Seoul, Republic of Korea


To reduce radiation dose and scan time, many research groups in the field of PET instrumentation are currently studying time-of-flight (TOF) technology which can improve the signal to noise ratio of the PET image. The scintillation light transport and collection plays an important role in improving the coincidence resolving time (CRT) of the PET detector based on the scintillators.

We assessed different scintillator array designs by investigating the impact of crystal pixel size and optical properties of the crystal (external reflector and surface treatment) on CRT for neuro TOF PET. One selected side surface was treated to be rough while the others polished. Four different crystal arrays were designed by considering array size, pixel size, reflector, and surface treatment: Design (1) 8 x 8 array, 3.11 x. 3.11 x 15 mm3, ESR, and one side-roughed; Design (2) 8 x 8 array, 3.11 x. 3.11 x 15 mm3, ESR, and entrance surface roughed; Design (3) 8 x 8 array, 3.19 x. 3.19 x 15 mm3, air-gap, and one side-roughed; Design (4) 12 x 12 array, 2.05 x. 2.05 x 15 mm3, ESR, and one side-roughed.

Comparing Design (1) and (2) to evaluate the effect of roughed surface location on CRT, both locations provided similar performance (about 263 ps). When the reflecting effect of air-gap and ESR was compared with Design (1) and (3), CRT using ESR was improved than air-gap by 12%. Finally, when Design (2) was compared to Design (4), CRT of Design (4) was reduced by 13% because of the declining light collection by 35%.

These experimental results showed that light collection and CRT were improved by using ESR than those by air-gap with the crystal array consisting of many pixels. Roughing at entrance surface was preferable than at side surface because CRTs were similar with the two treatments while the roughing entrance surface is easy to fabricate. The high spatial resolution PET detector for brain imaging having TOF capability could be achieved although some compromises in CRT are necessary.

Keywords: Time-of-flight, PET, LYSO Scintillator, Coincidence resolving time, Brain
Poster panel: 60

Poster Number:

X-ray phase-contrast simulations of fibrous phantoms using GATE (#1128)

J. Sanctorum1, J. De Beenhouwer1, J. Sijbers1

1 University of Antwerp, Department of Physics, Antwerpen, Belgium


The overall importance of x-ray phase-contrast imaging techniques has grown substantially due to their ability to generate contrast in a range of situations where traditional absorption imaging falls short. Monte-Carlo simulations are a valuable tool for benchmarking and testing new acquisition and reconstruction techniques, as well as optimization of the system and components design. Valuable efforts have already been put into x-ray phase contrast simulations, but so far little attention has been given to simulations intended for the fully three-dimensional reconstruction of the phase- and dark-field signal. In this work, we present a practical implementation of grating-based x-ray phase-contrast simulation in GATE, with specific focus on phantoms with fibrous microstructures. The simulations are based on a combination of Monte-Carlo simulations in GATE for modelling the x-ray interactions in the sample, and subsequent numerical wave propagation for describing the grating interformeter. We demonstrate the modelled physical processes that are involved in both parts of the simulation, and show simulated radiography and computed tomography results for the three imaging modalities that can be extracted from grating interferometry, namely absorption contrast, dark-field contrast and phase contrast.  The results show clearly the effect of scattering in fibrous samples on the different types of contrast.

Keywords: Phase contrast, Simulations
Poster panel: 63

Poster Number:

Simulation and Image Reconstruction of the Combined Siemens PET/CT and PET/MRI Systems  (#1239)

H. Kertész1, A. Renner2, I. Rausch1, T. Beyer1, J. Cal-Gonzalez1

1 Medical University of Vienna, QIMP group, Center for Medical Physics and Biomedical Engineering, Vienna, Wien, Austria
2 Medical University of Vienna, Digital Image Processing Laboratory, Center for Medical Physics and Biomedical Engineering, Vienna, Wien, Austria


The objective of this work is to validate a Monte Carlo (MC) simulation model for two commercially-available, whole-body PET systems. The MC models will be used to evaluate the performance of different image reconstruction methodologies at low count rates. 

GATE (GEANT4 Application for Tomographic Emission) was used as the MC toolkit for the modeling of the Siemens Biograph 64 TruePoint TrueView PET/CT (TPTV) and the Siemens Biograph PET/MR (mMR) systems. In both cases, we included detailed models of the detector electronics, system geometry and the physical processes involved in the data acquisition. The performance of both system models was validated following the NEMA (National Electrical Manufacturers Association) NU 2-2012 protocol. We compared the simulation results with the measured values for sensitivity, count rate (CR), and noise equivalent count rate (NECR). Moreover, three voxelized NEMA IQ phantom was simulated. The simulated data was reconstructed with the STIR framework using the standard OSEM algorithm. 

The calculated (reference value from measurements) sensitivity for the mMR was 13.8 (15.0) kcps/MBq and 14.4 (13.9) kcps/MBq at the center of the field-of-view (FOV) and at 10 cm radial offset, respectively. The NECR peak was 189 kcps @ 23.8 kBq/ml (184 kcps @ 23.0 kBq/ml) and the scatter fraction at the NECR peak was 29.0 (37.9) %. For the TPTV, the sensitivity was 8.0 (8.1) kcps/MBq and 7.9 (8.2) kcps/MBq at the centre of FOV and at 10 cm radial offset, respectively. The NECR peak was 151 kcps @ 27 kBq/ml (161 kcps @ 31 kBq/ml) and the scatter fraction at the NECR peak was 24.8 (32.5) %.   

Both PET/CT and PET/MRI models showed a good agreement (< 10 %) with the measured reference values. The application of these models for the evaluation of different image reconstruction algorithms in simulated numerical phantoms is work in progress.

Poster panel: 66

Poster Number:

Simulations of RF screen topologies for a shared-volume PET insert for 7T small-bore MRI (#1437)

L. Yin1, V. Schulz1

1 RWTH Aachen University, Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, Aachen, North Rhine-Westphalia, Germany


We evaluated radio frequency (RF) screen topologies for a positron emission tomography/magnetic resonance imaging (PET/MRI) insert at 7T. Conventional small-animal PET/MRI inserts underlie strict space limitations, as the usable space inside the MRI bore is small. Passing the scintillators through the screen into the RF volume (i.e., shared-volume) overcomes this limit: While enabling a closer module placement, a higher radius for the RF screen can be used and thus both systems' sensitivities are enhanced. However, the field penetrating the openings may create mutual RF interferences which need to be limited. We selected two geometries for this task: Cylindrical and polygonal prismatic screens with varying radii were simulated in CST Microwave Studio to investigate the resulting influence on the coil sensitivity (CS) and shielding effectiveness (SE) at 298MHz. The SE was monitored along several positions close to the PET sensors.

We investigated closed and modified screens at five radii from 28 to 43.5mm. Openings were created at the scintillator positions. Additional shielding structures were added to the openings for defined radii to potentially increase the SE while maintaining a good CS.

A screen radius of 35.75mm yielded the best trade-off between CS and SE. The CS is normalized using the maximal achievable value for a closed screen at r=52mm and reached more than 87% for both geometries. The SE was higher than 31dB. Setting the screen radius close to the sensor level, i.e. to 43.5mm, increases the field near the openings' edges and should be avoided without additional shielding. Nevertheless, adding components to an opened polygonal prismatic screen at 43.5mm proved that increasing the SE while maintaining a sufficient CS is possible: Of all the components tested, an extension of the openings' edges towards the resonator showed the best results. A CSnorm of 93.437% and SE between 20 and 24dB were reached depending on the monitored position and extension depth.

Keywords: RF screen, Simulation, Shared-volume, 7T PET/MRI, RF shielding
Poster panel: 69

Poster Number:

Optimization of a High Resolution Small Animal SPECT System using GATE and STIR Software (#1776)

H. G. Kang1, H. Tashima1, S. J. Hong2, 3, 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 Eulji University, Department of Radiological Science, Seongnam, Republic of Korea
3 Eulji University, Department of Senior Healthcare, Daejeon, Republic of Korea


Single photon emission computed tomography (SPECT) has been widely used for preclinical research. High resolution small animal SPECT imaging can be achieved by employing a pinhole or multi-pinhole collimators. However, the useful FOV is narrow compared to that of a parallel-hole collimator. The aim of this study is to optimize the hole diameter of a parallel square-hole collimator for high resolution whole-body small animal SPECT imaging using GATE and STIR software. The small animal SPECT consists of a tungsten parallel square-hole collimator (L= 30 mm), a monolithic GAGG crystal (49 × 49 × 5 mm3), and a PMT (H9500). The SPECT images of an Ultra-Micro hot phantom and NEMA NU-4 phantom were acquired with different square-hole sizes of 0.2, 0.4, 0.6, 0.8, 1.0, and 1.2 mm. The collimator length and septal thickness were kept as 30 mm and 0.1 mm, respectively. The projection image (128 × 128 matrix, pixel size of 0.3828 × 0.3828 mm2) was reconstructed with FBP, and OSMAPOSL algorithms provided by STIR software. With the collimator hole size of 0.4 mm, the spatial resolution of 1.11 mm and sensitivity of 0.001% could be achieved. The contrast recovery coefficient could be improved by using the 0.4 mm hole size and the hot rods (D= 1.35 mm) could be identified clearly. The important message from this work is that the small animal SPECT system can be optimized with the open-source software packages of GATE and STIR. In the future, the optimal septal thickness will be investigated to improve the sensitivity without compromising the spatial resolution.

Keywords: GATE, STIR, SPECT, collimator
Poster panel: 72

Poster Number:

Simulation and design of a New Plug&Imaging sensor for novel Digital brain PET system   (#1842)

M. Gao1, Q. Xie1, 2, E. Antonecchia1, 2, Y. Hua3, D. Xi4, C. - M. Kao5, N. D’Ascenzo1, 2

1 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
2 NEUROMED I.R.C.S.S, Pozzilli, Italy
3 Raycan Technology Co., Ltd., Suzhou, China
4 RayMeasure Medical Technology Co., Ltd., Suzhou, China
5 The University of Chicago, Department of Radiology, Chicago, United States of America


Positron Emission Tomography (PET) is one of the most  accurate techniques currently used to provide a diagnosis of neurological diseases, such as Alzheimer, Parkinson, Lateral Sclerosis and malignant neoplasms. PET retrieves the space distribution of the  positron emitting ligands, which are  developed  in order to target the specific disease under investigation.
Our group has previously proposed a digital multi voltage threshold (MVT) sampling method that takes samples of a pulse with respect to a set of four reference voltages. The total energy and the timing of the pulse can be obtained by the digital signal processing techniques. The combination of novel LYSO/SiPM sensor technology, MVT dedicated digital electronics and image reconstruction constitutes the Plug\&Imaging (P\&I) sensor.
In this paper we present for the first time a simulation framework based on GEANT4 and including all the characteristic features of the  digital Plug and Imaging (P\&I) sensor system for Positron Emission Tomography. We manufactured a prototype of Digital Brain Positron Emission Tomography and we compared the experimental results with the simulation model in terms of  sensitivity and  space resolution estimation. We obtain an expected sensitivity of approximately 10~cps/kBq and a space resolution of approximately 1 mm.

Keywords: simulation, Brain PET
Poster panel: 75

Poster Number:

Deep Neural Network for Obtaining Bone-suppressed Radiographs (#2020)

E. Park3, Y. Ha1, S. Ha1, H. K. Kim1, 2

1 Pusan National University, School of Mechanical Engineering, Busan, Republic of Korea
2 Pusan National University, Center for Advanced Medical Engineering Research, Busan, Republic of Korea
3 Hanwha Defense Systems, Launching System R&D center, Changwon, Republic of Korea


This paper describes the deep neural network (DNN) for a bone-suppressed chest radiograph. For the training of the DNN, numerous regions of interest (ROI) are selected from the conventional chest radiograph as the training data and the center values of the ROIs are extracted from the bone-enhanced chest radiograph as the teaching data. The number of training and teaching dataset is 6,000 and the size of ROI is 21x21. For the solution to the vanishing gradient problem, the rectified linear unit is selected as the non-linear activation function. In addition, the dropout technique is applied to the DNN to cope with the overfitting problem. The conventional and bone-enhanced chest radiographs are obtained by a dual-shot DEI system. The original chest radiograph has the size of 2022x2022, but for computational efficiency, the size is shrunk to 400x400. The bone-suppressed chest radiograph is obtained by the weighted log-subtraction using the conventional and the bone-enhanced one using the DNN. The acquisition of the bone-suppressed chest radiograph will be conducted and the comparison of the bone-suppressed image by the DEI system and the DNN.

Keywords: deep neural network, bone-suppressed radiograph, chest radiograph
Poster panel: 78

Poster Number:

A concept of monochromatic X-ray beam with K-edge filter for low dose mammography: A simulation study (#2126)

C. - H. Baek1, D. Kim2

1 Kangwon National University, Samcheok, Republic of Korea
2 Eulji University, Seongnam, Republic of Korea


X-ray mammography may acquire an image for early breast cancer detection in medical application. There have been a number of approaches in the low-dose mammography. The purpose of this study was to design a concept of monochromatic X-ray beams by using K-edge filter for low dose mammography. The X-ray simulation code of SRS-78 and Monte Carlo simulation code of GATE were used to model the X-ray tube, target material, filter material, breast phantom, and detector. To achieve monochromatic X-ray beam, molybdenum (Mo) and a rhodium (Rh) is used as additional filters material with various thicknesses. The direct conversion detector(FDXD 1417, Drtech, Seongnam, Korea) composed of thin-film transistor (TFT)-amorphous selenium (a-Se) was modelled through Monte Carlo simulation study.

According to the CVR and EER results, filter thickness for mammographic imaging is determined to 6 and 3 HVL for Mo and Rh, respectively. The improvement of SNR values with Rh filter compared to that without of filter is 1.01, 6.28, 5.60, 5.60, 5.60, 5.60 % from 0.1 and 0.6 μGy. The present work demonstrates that monochromatic X-ray beams for low dose mammography. According to the results, Rh filter could be useful for enhancing calcification while absorbed dose is reduced.

Keywords: mammography, low dose, filter
Poster panel: 81

Poster Number:

Modelling positron tracks in biologically relevant media: application to F-18 (#2245)

G. Garcia1

1 Consejo Superior de Investigaciones Científicas, Instituto de Física Fundamental, Madrid, Spain


In this study our Low Energy Particle Track Simulation code has been used to simulate single positron tracks in liquid water and pyridine. The input parameters used for the Monte Carlo simulation are the positron scattering cross section for all possible interaction processes with the target molecules in the energy range (0-600 keV). This data set is derived from our previous measurements and calculation completed with available data in the literature. These processes include elastic scattering, electronic, vibrational and rotational excitation, ionisation, positronium formation and direct annihilation as well as those corresponding to the generated low energy secondary electrons. This representation provides information about the energy deposition in selected volumes as well as the number and type of interactions taking place in that volume. This level of detail allows to characterise the radiation damage in terms of energy deposition, as traditional dosimetry does, and also to correlate this damage with induced molecular dissociations, i.e. describing radiation damage at the molecular level (nanodosimetry). The present modelling procedure is applied to the case of F-18, a positron emitter commonly used in PET scanning diagnostics. The emission spectrum has been determined with a silicon spectrometer and used to represent the primary positron source for the simulation.

Keywords: Positron interactions, Monte Carlo modelling, Positron transport
Poster panel: 84

Poster Number:

Simulation of a High-Sensitivity Adjustable-FOV PET Scanner (#2315)

I. Ozsahin1, M. S. Musa1

1 Near East University, Department of Biomedical Engineering, Nicosia, Turkey


The aim of this study is to evaluate a new Positron Emission Tomography (PET) geometry, whose target was to improve the overall system sensitivity and design a patient/disease specific adjustable FOV. We proposed two PET system designs, a sandwich-PET and a multi-purpose-PET which were simulated using Geant4 based Architecture for Medicine Oriented Simulations (GAMOS) software. They are made to have a similar geometrical configuration, except that the multi-purpose scanner is one-third of the sandwich-PET scanner in the axial direction. We call the first design "sandwich" because the system with the size of 66.2 cm and 193.2 cm on the transaxial and axial direction, respectively, resembles a sandwich when the patient is between the detector heads. The system is intended for whole-body studies, whereas the multi-purpose system which has a total size of 66.2 cm on both the axial and transaxial direction can be used for brain and cardiac studies, as well as cancer imaging. Both system’s transaxial field of view (FOV) can be adjusted depending on the patient type and body region to be examined, and they employ lutetium yttrium oxyorthosilicate (LYSO) crystals with 2.76 2.7618.1 mm3 pixel sizes. The performance evaluation test was conducted using a simulated phantom with a point source placed at several locations on the axial FOV. The coincidence events were detected with an energy window of 350–650 keV, and coincidence time window of 20 ns. The result showed that the system’s sensitivity increases exponentially when varying the transaxial FOV from 70 cm (regular openings for conventional PET scanners for all type of patients: slim, overweight, or obese) down to 20 cm (for slim or pediatric patients). This study shows that the system is capable of achieving much higher sensitivity comparing to the conventional whole-body scanners.

Keywords: FOV PET Scanner
Poster panel: 87

Poster Number:

Sensitivity Estimation and Image Reconstruction for Sparse PET with Deep Learning (#2408)

T. Feng1, J. Wang1, H. Li1

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


The use of a sparse crystal setting would reduce the cost of the PET scanner and has advantages such as less RF shielding in PET/MR. It also allows a longer axial field of view (FOV) using the same crystal volume.

In this paper, the sensitivities of the coincidence events of PET systems with the sparse setting, thin crystal setting, and the conventional design using a fixed total crystal volume were analytically estimated. The sinograms of a sparse system (with 50% crystal removed and fixed axial FOV) were simulated using patient data. Reconstruction algorithms were developed by modeling the effects of reduced crystals in the system matrix. A convolutional neural network (CNN) based noise reduction approach was used for post-processing. A total of 14 patient data were included and were truncated to 3 minutes scan for consistency. Leave-one-out cross-validation was used for evaluation purpose. A patch based data input/output were used for model training to increase the number of training samples. Images reconstructed using OSEM followed by Gaussian denoising was also used as comparison. The percentage summed square difference (SSD) between images of sparse setting and non-sparse systems were used for quantitative evaluation.

When using the same total volume of crystals, the difference of sensitivity at the center of FOV was within 10% among three different settings, with the rank from highest to lowest being the thin detector, sparse detector, and conventional detector. When using the same axial FOV, reconstructed images of the sparse setting showed increased noise due to reduced sensitivity. The percentage SSD for image processed with the Gaussian filter was 30% on average and was reduced to 16% with CNN on average.

The results show with the same amount of crystal, the use of sparse setting provides a slightly larger sensitivity and much larger axial FOV. CNN processed images was able to partially recover lost image quality due to the removal of certain crystals.

Keywords: Sparse PET, Image reconstruction, deep learning
Poster panel: 90

Poster Number:

Development of an EDUGATE simulation toolkit for the easyPET scanners family (#2438)

P. M. M. Correia1, J. Menoita1, N. O. Romanyshyn1, A. L. M. Silva1, J. F. C. D. A. Veloso1

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


EasyPET is a new concept of a Positron Emission Tomography (PET) scanner using an innovative acquisition method based on two rotation axes for the movement of detector pairs. Due to its simplicity, its suitable for education purposes, to teach students about the PET technology and its basic concepts, from the radiation detecting and analog pulse analysis to the coincidence sorting and image reconstruction. The concept allows achieving high and uniform position resolution over the whole field of view (FoV), by eliminating parallax errors due to the depth of interaction (DoI), which are typical of ring-based PET systems, so quality images are obtained even without state-of-the-art image reconstruction algorithms. The technology developed at the University of Aveiro with a patent-pending, is licensed to CAEN S.p.A, and included in the educational catalog of the company. In this work, a simulation toolkit based in the easyPET available from CAEN is presented. It can simulate all the physical aspects of the product, such us the scanning range, variable Field-of-View (FOV), scintillator energy resolution, coincidence time and energy window, among others. A simple image reconstruction algorithm based on Filtered-back-projection (FBP) is implemented. The toolkit allows a quick analysis in classroom of the simulation results. The platform was also used to study the new EasyPET 3D version, and a simulation of a NEMA NU 4-2008 IQ phantom was performed, demonstrating the capability of the platform  not only for education purposes but also for research.


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: PET, Edugate, Simulation, EasyPET
Poster panel: 93

Poster Number:

Fast gamma-ray event interaction-position-estimation using k-d tree search  (#2529)

X. Li1, L. Tao2, C. S. Levin2, L. R. Furenlid1

1 University of Arizona, Radiology Research Laboratory, Tucson, Arizona, United States of America
2 Stanford University, Department of Radiology, Stanford, California, United States of America


We have developed a fast gamma-ray interaction position estimation method using k-d tree search. Compared with traditional methods, this method can achieve both speed and accuracy at the same time using k-d tree data structure. The k-d tree search method has a time complexity of O(log2(N)), where N is the number of entries in the reference dataset, which means larger reference datasets can be used to efficiently estimate each event’s interaction position. This method’s accuracy was found to be equal to that of the exhaustive search method, yielding the highest achievable accuracy. Most importantly, this method has no restriction on the data structure of the reference dataset and can still work with complicated mean-detector-response functions (MDRFs), meaning it is more robust compared with other methods such as contracting-grid (CG) search or vector-search (VS) methods. Estimation of 9.3×10^5 events for the edge-readout detector of dimension 50.8mm × 50.8mm, exhaustive search cost 4292 secs, CG search cost 26 secs while k-d tree search cost only 14 secs. If using the result of exhaustive search as ground truth, CG search has a standard deviation of 1.3 mm, while k-d tree search has a standard deviation of exactly 0 mm, meaning it provides the same precision as exhaustive search method. 

Keywords: gamma-ray, position estimation, k-d tree, scintillator detector
Poster panel: 96

Poster Number:

GPU-Assisted Generation of System Matrices for High-Resolution Imaging Systems (#2643)

A. L. Lin1, X. Li1, L. R. Furenlid1, M. Kupinski1

1 University of Arizona, College of Optical Sciences, Tucson, Arizona, United States of America


We present a method for the fast generation of point spread functions (PSF) used in the creation of system matrices for gamma-ray imaging systems incorporating high-resolution detectors such as those based on semiconductors.  This algorithm builds upon the methodology introduced in a previous GPU-based ray-racing method where the intrinsic detector resolution was a significant source of blur.  In contrast to that algorithm, which traced photons all originating from an idealized point source located at the center of a voxel, our new approach stochastically samples a finite-sized point source volume that accurately models experimental PSF measurements.  This change allows for an efficient and realistic simulation of how continuous objects are translated through the imaging system when imaged with the new generation of solid-state pixel and crossed-strip detectors that have submillimeter intrinsic spatial resolution.  We investigate the benefits and limitations as a function of pinhole and detector configuration, experimental point source dimension, and assess the impact on voxel interpolation.

Keywords: System Matrix Generation, GPU
Poster panel: 99

Poster Number:

Accurate method providing intrinsic spatial resolution on monolithic based PET detectors  (#2716)

A. González-Montoro1, F. Sánchez1, P. Bruyndonckx2, G. Cañizares1, J. M. Benlloch1, A. J. González1

1 Universitat Politècnica de València, Institute for Instrumentation in Molecular Image (i3M) -CSIC, Valencia, Spain
2 Bruker BioSpin, Kontich, Belgium


The aim of this work is to provide a method to retrieve the intrinsic resolution of detector blocks based on monolithic crystals but in the assembled scanner. Estimating the intrinsic spatial resolution is typically challenging. This resolution is limited by several factors such as detector size or photon depth of interaction in the scintillator. These factors cannot easily be isolated and studied experimentally, for this reason many works focused on exploring the resolution limits through Monte Carlo simulations or other models. 

The proposed method can be done in-situ, and it has been validated against a more traditional approach of using bench-top set-up where a detector for collimation is moved backwards. We have tested two different detector configurations[1], namely, (i) trapezoidal block with dimensions of 50 × 50 mm2 base (exit), 47.9 × 47.9 mm2 entrance face and 10 mm thickness, and (ii) rectangular block with dimensions of 50 × 50 mm2 and 20 mm thickness. The entrance face of the detector under study is coupled to a retro-reflector (RR) layer. and the exit face to an array of 16×16 SiPMs. A third configuration with a rectangular block of 15 mm thickness is now being studied.

A mathematical model has been derived to fit the experimental software collimated profiles.  The model accounts for the combined influence of the Gaussian intrinsic resolution PSF, the 1D Lorentzian source distribution obtained from MC simulations and the geometrical effect of the software collimation.  The free parameters in the fit are the width of the intrinsic resolution PSF and the average interaction depth.

The experiments resulted in a detector intrinsic spatial resolution of 0.83 mm FWHM for the 10 mm thick crystal, and degrades to 1.29 mm FWHM in case of the 20 mm thick block. These tests show a accurate method to determine the intrinsic resolution of monolithic-based detector blocks in a fully assembled PET system.

Keywords: gamma ray detectors, photodetectors, positron emission tomography, solid scintillation detectors, intrinsic resolution
Poster panel: 102

Poster Number:

Design and Initial Performance Evaluation of a Large-Animal Dynamic SPECT/CT Imaging System (#2794)

Z. Lyu1, 2, H. Wang1, 2, H. Liu3, L. Cheng1, 2, Y. Liu1, 2, S. Wang1, 2, Z. Wu1, 2, T. Ma1, 2

1 Tsinghua University, Department of Engineering Physics, Beijing, China
2 Ministry of Education (Tsinghua University), Key Laboratory of Particle & Radiation Imaging, Beijing, China
3 Yale University, Department of Internal Medicine(Cardiology), New Haven, Connecticut, United States of America


Objective: Comparing with small animals, large animal models are more suitable for a variety of translational research. For large animal, existing SPECT systems are less optimal, due to limited FOV, spatial resolution and/or sensitivity. In this work, we aim at developing a large-animal dynamic SPECT/CT system with multi-pinhole collimators and a novel slit-slat&multi-pinhole collimator.
Methods: The large animal SPECT consists of six stationary clinical SPECT detectors arranged in hexagon (radius: 470mm). The large FOV collimator consists of six 2×4 pinhole groups whose position and opening angle are designed to cover whole FOV (Φ200mm×300 mm). A novel slit-slat&multi-pinhole collimator is designed to achieve different performance in different region of FOV. The collimator consists of three slit-slat (for body FOV) and three multi-pinhole (for heart FOV). Each multi-pinhole includes 4×4 pinholes (diameter: 4.4mm; open angle: 29°). The slit width is 5.8mm and open angle is 98°. The slat gap is 1.92mm and thickness is 0.2mm. The theoretical calculation and simulation are performed to evaluate performance of different collimators.
Result: For large FOV collimator, spatial resolution and sensitivity over pinhole diameter curves are calculated and pinhole diameter is determined to 2.7mm. In the center of FOV, spatial resolution is 5.9 mm and sensitivity is 0.0393%. For slit-slat & multi-pinhole collimator, spatial resolution and sensitivity are 8.0mm, 0.12%(heart FOV) and 10.6mm, 0.013%(body FOV). In two hotrod phantoms (3.0~5.5mm and 5~10mm) simulation, 5.0mm hotrod can be distinguished.
Summary: The proposed system has capability to realize large animal dynamic imaging. The multi-pinhole collimators and slit-slat&multi-pinhole collimator were designed and performance were evaluated by calculation and simulation. The prototype system with two detector will be accomplished to perform preliminary experiment.

Keywords: SPECT, large animal, dynamic, novel collimator
Poster panel: 105

Poster Number:

Assessment of easyPET for 18F-FDG neurological studies in mice (#1248)

F. M. Ribeiro1, A. L. M. Silva1, P. M. M. Correia1, I. F. Castro2, P. M. C. C. Encarnação2, F. M. Rodrigues2, A. C. Santos3, C. Ramos3, N. C. Ferreira3, 4, D. A. Sá3, 4, D. G. Priolli5, C. Nicolucci5, J. F. C. D. A. Veloso1

1 University of Aveiro, i3N - Department of Physics, Aveiro, Portugal
2 RI-TE - Radiation Imaging Technologies, Lda, UA Incubator, Ílhavo, Portugal
3 Faculty of Medicine, University of Coimbra, iCBR - Institute for Clinical and Biomedical Research, Area of Environment Genetics and Oncobiology (CIMAGO), Biophysics Institute, Coimbra, Portugal
4 University of Coimbra, ICNAS, Coimbra, Portugal
5 São Francisco University, Multidisciplinary Research Laboratory, Bragança Paulista, Brazil

easyPET collaboration


Preclinical Positron Emission Tomography (PET) systems can be applied for small animal imaging to study human diseases, to validate new drugs and therapeutics, as well as to develop new PET radiopharmaceuticals for diagnosis. However, the access to preclinical PET scanners is not easy due to its high complexity and cost. EasyPET, a new concept of PET scanner with a simple and unique image acquisition method based on two rotation shafts for the movement of detector pairs, represents a solution to this problem, while keeping a fair sensitivity and achieving state-of-the-art spatial resolution. Small-animal imaging has mainly been applied in the field of neurology for preclinical evaluation of potential therapies. In this work, the first results obtained with a small-scaled version of easyPET (training version) for mice brain imaging with 18F-FDG are presented. The subject received different stimuli, which activated in a specific way the brain regions. An acquisition without any stimulus was used as control. The results suggest that easyPET is suitable for preclinical imaging research, since it produces high quality mice images, by clearly identifying the main brain regions.
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: PET, preclinical, neurology
Poster panel: 108

Poster Number:

Performance of A Long Rectangle Semi-Monolithic LYSO Scintillator PET Detector (#1465)

X. Zhang1, 2, X. Wang1, N. Ren1, Z. Kuang1, X. Fu1, C. Zhang1, S. Wu1, Z. Sang1, Y. Yang1

1 Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Institute of Biomedical and Health Engineering, Shenzhen , China
2 Lanzhou University, School of Nuclear Science and Technology, Lanzhou, China


As compared to monolithic scintillator PET detector, the semi-monolithic detector has advantages of reducing edge effects and simplifying the calibration process. In this work, the performance of long rectangle semi-monolithic scintillation detectors with two different crystal face treatments was measured, and different positioning methods were investigated. The semi-monolithic scintillator consists of 11 polished LYSO slices measuring 1×37.6×10 mm3. The detectors were read out by a 4 ×12 SiPM array with a 1.5 mm light guide and silicon grease in between them. First, the detector was uniformly irradiated with a point 22Na source, and slice identification maps and the energy spectrum of each slice were measured. Second, the detector was irradiated at a series of selected positions (y, z) by electronic collimation using a 0.25 mm diameter 22Na source and a single 1×1×20 mm3 LYSO crystal detector. The traditional center of gravity (COG) method and the squared COG method were used for y position estimation, and three different methods were used for z position estimation. The measured y linearly changes with the true y position until ~ 5 mm to both ends of the detectors. Both the y and z position resolutions degrade as the positions are farther from the SiPMs. Compared to the COG method, the squared COG method provides better y position resolutions and smaller positioning bias at both ends. Painting of the crystal faces largely reduces edge effects and improves both y and z position resolutions, but degrades the energy resolution and timing resolution.

Keywords: PET; Semi-monolithic scintillator; Depth of interaction; SiPM
Poster panel: 111

Poster Number:

Design of a selectable pinhole module for use in adaptive SPECT collimators (#1619)

N. C. Momsen1, G. Richards1, G. I. Zubal2, M. A. King3, L. R. Furenlid1, 4

1 University of Arizona, College of Optical Sciences, Tucson, Arizona, United States of America
2 Z-Concepts LLC., East Haven, Connecticut, United States of America
3 University of Massachussetts, Department of Radiology, Worcester, Massachusetts, United States of America
4 University of Arizona, Department of Medical Imaging, Tucson, Arizona, United States of America


We present a module for the real-time control of gamma-camera pinhole apertures. The mechanical design includes the ability to rapidly select between an array of pinhole diameters and pinhole states (whether open or shuttered). This is achieved through the use of a wheel with multiple pinhole bores configured with the same entrance angles. Control of the module is via wireless Arduinos using either Bluetooth or Wi-Fi to avoid wiring trailing from the system. Modulation of the aperture can either be done via manual control by the user, or through an automated routine integrated into the data acquisition code. Because we can control pinhole diameter, as well as the number and position of pinholes that are open in a multi-pinhole configuration, system resolution and sensitivity are dynamic variables that we can change for each individual acquisition, or within an individual scan (based on a priori information or a scout scan). The integration of these modules into the gamma ray camera is discussed, including the drive mechanism used to achieve the rotary motion and an opto-interruptor module used to obtain feedback for reproducible positioning. Two applications are discussed: apertures for clinical, commercial gamma cameras, and apertures for a dedicated brain imaging system.

Keywords: SPECT, Pinhole, Adaptive SPECT
Poster panel: 114

Poster Number:

In-vivo Coronary Micro-CT of Small Animals (#1598)

J. Kuntz1, 2, C. Funck1, 2, J. Maier1, 3, M. Kachelrieß1, 2, S. Sawall1, 2

1 German Cancer Research Center (DKFZ), X-Ray Imaging and CT (E025) , Heidelberg, Baden-Württemberg, Germany
2 Ruprecht-Karls-University, Medical Faculty, Heidelberg, Baden-Württemberg, Germany
3 Ruprecht-Karls-University, Department of Physics and Astronomy, Heidelberg, Baden-Württemberg, Germany


Cardiovascular diseases are the leading cause of death worldwide and major topics in clinical and preclinical research. However, image quality in current preclinical micro-CTs suffers from limited temporal and spatial resolution and thus is not adequate for the in-vivo imaging of coronary arteries in the used murine animal models. In this work, the first results of a prototype CT system dedicated to murine coronary artery imaging are presented. In total, four mice were examined after administration of a blood pool contrast agent. Phase-correlated images were reconstructed into 10 equally distributed cardiac phases using the Feldkamp-algorithm. In all individuals, the left anterior descending coronary artery is visible in all reconstructed cardiac phases. The measured contrast of 370 HU is sufficient to determine functional parameters like ejection fractions or for the measurements of myocardial wall thickness. The presented in-vivo coronary micro-CT technique can emerge as a powerful tool for preclinical research.

Keywords: Small animal, micro-ct, computed tomography, cardiac imaging
Poster panel: 117

Poster Number:

Low-dose, high-sensitivity Positron Emission Mammography (PEM) system for screening of high-risk population (#1933)

O. Bubon1, 3, H. Poladyan1, V. Zavarzin2, E. Anashkin3, A. Reznik1

1 Lakehead University, Physics Department, Thunder Bay, Ontario, Canada
2 Analog & Digital Precision, LLC, Boston, Massachusetts, United States of America
3 Radialis Medical Inc., Thunder Bay, Ontario, Canada


Mammographic density has been recognized as the greatest contributor to the failure of screening x-ray mammography in high-risk women. Dense breast tissue looks similar to malignancy on mammograms, leading to inconclusive imaging results. Fully solid-state Positron Emission Mammography (PEM) imaging system developed in our group improves specificity and sensitivity for breast cancer detection in both diagnostic and high-risk screening settings, alleviating the uncertainty from imaging. The improvement was made possible by the development of a new type of four-side tileable sensor modules, which are seamlessly combined together – without gaps or dead zones – into a sensor area of a needed size to fully cover the entire breast.

Here we report on evaluation of a small field-of-view (FOV) PEM prototype where 8 individual sensor modules working in coincidence were seamlessly connected in 2D array to form an imaging area of 12 by 12 cm2.  Each module employs silicon photomultiplier (SiPM) of 64 pixels coupled to LYSO crystal arrays. Custom-made 16 channel detector acquisition system coupled to a 4-channel multiplexing preamplifier was used to read out and process coincidence signal from 8 detector modules. Per NEMA guidelines spatial resolution was evaluated across full FOV in all three planes using Na-22 point source. Measured system spatial in-plane (X,Y planes) resolution is 1.5-2.2 mm with degradation at the edges of FOV. Measured axial (Z-plane) spatial resolution ranges from 4.5 to 6 mm depending on the detector module separation. PEM system imaging performance was evaluated using NEMA small animal PET phantom filled with F-18 radioisotope. Evaluation of sensitivity measurements showed that absolute sensitivity of 1.86% is improved by 12 times to compare to commercially available PEM imaging systems allowing low-dose PEM imaging. Sensitivity and spatial resolution were evaluated for various detector heads separations and different event acceptance angles.

Keywords: Positron Emission Mammography, SiPM, low-dose imaging, Molecular Breast Imaging, PET
Poster panel: 120

Poster Number:

Elimination of Inter Crystal Scattering Events in the X'tal Cube PET Detector with (0.77 mm)3 Segments (#2039)

M. Nitta1, 2, F. Nishikido1, N. Inadama1, H. Tashima1, H. Kawai2, T. Yamaya1

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


 We have developed the X’tal cube PET detector which ‘ is composed of 96 multi pixel photon counters (MPPCs) and a (13.1 mm)3 LYSO scintillator segmented into the 17 x 17 x 17 array. Each segment size is (0.77 mm)3. Six 4 x 4 MPPC arrays cover all six surfaces of the LYSO scintillator. The Anger calculation from 96 MPPC signals is used to determine the segment detecting a gamma-ray. We confirmed that the X’tal cube achieved sub-millimeter isotropic spatial resolution. However, we also observed inter crystal scattering (ICS) events as background events. There is a concern that ICS events degrade contrast of PET images for the high spatial resolution PET scanner.

 We proposed a method to discriminate the ICS events and applied it to the X’tal cube. If the photo absorption event occurs at one segment, the distribution of spread light on each MPPC is the same for every event. On the other hand, even if the result of the Anger calculation is the same as the photo absorption event, the distribution of spread light by the ICS events differs from that of the photo absorption event because scintillation light is generated at different segments. To discriminate the ICS events,  first using a uniform irradiation experiment, we obtained the mean and standard deviation of pulse height peak of each MPPC for the segment to set the threshold for the discrimination of the ICS events. Then for each detected event, each MPPC pulse height was measured and the chi-square value was calculated. If the chi-square value exceeded the threshold, the event was eliminated as an ICS event. Next, to prove that our method can eliminate the ICS events, we obtained the response function by a fan-beam scanning experiment. We compared the results before and after applying our method. At the central part of the X’tal cube, full width half maximum of the ICS distribution in the response function was improved from 4.0 mm to 2.1 mm and over 50% of the ICS events were eliminated.

Keywords: sub-millimeter spatial resolution, PET detector, inter crystal scattering, the X'tal cube, DOI detector
Poster panel: 123

Poster Number:

Simulation, design, fabrication and assessment of a wearable brain PET camera (#2068)

Z. Zhao1, F. Weng1, S. Xie2, Q. Huang1, J. Xu2, Q. Peng3

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


PET is a powerful tool in both the neurologic studies and brain-related clinic applications. The conventional method of imaging subject in supine is not always desirable for brain imaging.

In this study, we have designed and fabricated a wearable PET camera for brain imaging. The aperture and the AFOV of the system are 200 mm and 33.6 mm, respectively. The PET camera has 16 detector modules, 16 readout electronics boards with 100 channels and a 3D printed gantry. The detector module consists of a 10 × 10 LYSO crystal array (single crystal size: 3 mm × 3 mm × 20 mm), and a 10 × 10 SiPM array. The discrete crystals and the SiPMs are one-to-one coupled. The total weight of the scanner is 3,325 grams. We also proposed the Hyperloop data link (HDL) to solve the problem of data transmission between detector modules. In addition, we have simulated six different designs with the same weight of LYSO scintillators to fully compare the imaging performance.

We are currently performing more phantom studies to assess its performances. This wearable brain PET camera makes it possible to monitor the neurologic activities in the brain when the subject in more natural states, such as sitting and walking.

Keywords: Wearable brain PET, Detector, Readout electronic circuit, phantom
Poster panel: 126

Poster Number:

Plant-Specific Modular PET: Data Processing with CASToR and Performance Evaluation (#2314)

Y. - F. Chang1, A. Talebitaher1, K. R. Thompson1, Z. Papandreou1, S. Siciliano2, A. Teymurazyan1

1 University of Regina, Department of Physics, Regina, Saskatchewan, Canada
2 University of Saskatchewan, Department of Soil Science, Saskatoon, Saskatchewan, Canada


A modular PET detector system with variable geometry, BioPET, has been constructed to specifically study plant-microorganism-environment complexes. While the modular design in combination with rotation and translation mechanics provides greater flexibility in detector arrangement, adaptive data processing and image reconstruction methods are needed for such a transformable system. In this work, we developed a data processing pipeline by making used of a generic, modular, and extensible platform for tomography reconstruction, CASToR, and applied the process to evaluate performance of BioPET in rotational and stationary modes. The NEMA NU 4-2008 image quality phantom was utilized for analysis of recovery coefficients, spillover ratios, uniformity, and their relationship to selected reconstruction parameters. The rotational mode outperforms the stationary mode in exhibiting lower spillover effect as well as in reconstructing the activity concentration of smaller features, although partial volume effect is evidenced in both modes. The behaviour of the performance indices as a function of the selected reconstruction parameters is found specific to each detection mode. The procedure and the image quality characterization presented in this work are driving the development and optimization of data processing, image reconstruction and correction framework towards quantitative image analysis for PET systems with flexible geometry.

Keywords: Data Processing, PET
Poster panel: 129

Poster Number:

Calibrations of the Integrated Circuit Readout for a High Resolution Preclinical CZT PET Imaging System (#2522)

A. Groll1, C. S. Levin1

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


This work focuses on the development of a software control scheme which is capable of (A) automated ASIC calibration, (B) per channel analog readout electronics programming, and (C) executing user imaging protocols for a preclinical sub 1-mm spatial resolution CZT PET system. The system is built from 24 dual-CZT detector modules; each CZT crystal is 40 x 40 x 5 mm3 and the electrodes deposited on each crystal in the dual module are arranged back-to-back in an anode-cathode-cathode-anode configuration. Each CZT crystal has 39 anode strips, and 38 electrodes. Anode strips are 100 μm width with a 1 mm pitch and steering electrode has 400 μm width. The cathode side of our detector module is composed of 8 strips which are 4.9 mm width with a pitch of 5 mm. Crystals are configured in an “edge-on” orientation in which photons enter the 5x40 mm2 and pass through a much thicker CZT volume resulting in a detection efficiency equivalent to that of 2 cm of LSO scintillator. The cathode and anode strips are supported by a custom front end electronics board designed around the RENA-3 (Readout Electronics for Nuclear Applications) ASIC. A sub-assembly of 2 x 18 RENA boards has been assembled and currently contains a total of 2592 channels. The full system will have 4x48 RENA boards which is supported by 13,824 channels.  Given the need to set preamplifier, amplifier, and triggering parameters per channel, a robust operations infrastructure is necessary to program, and operate each RENA 3 board. We have developed an iterative, multi-channel programming and DAC threshold calibration scheme and performed preliminary testing on our CZT PET sub-assembly. Using a subset of detectors, a total of 432 channels were programmed and calibrated. Programming and calibration required a mean time of 5.17 minutes per ASIC with a standard deviation of 0.39 minutes. Spectra were acquired to validate the DAC triggering threshold of fast and slow RENA 3 channels. 

Keywords: Pet, preclinical, CZT, Small Animal, High Resolution
Poster panel: 132

Poster Number:

Non-destructive Characterization of Internal Structure of Crowned Teeth by Neutron Imaging (#2655)

A. Tremsin1, T. Shinohara2, K. Oikawa2, W. Kockelmann3, T. Minniti3, F. Salvemini4, T. Smithers5

1 University of California, Berkeley, Space Sciences Laboratory, Berkeley, California, United States of America
2 Japan Atomic Energy Agency, Naka-gun Ibaraki , Japan
3 Rutherford Appleton Laboratory, Harwell, United Kingdom
4 Australian Centre for Neutron Scattering, Lucas Heights, Australia
5 T.E. Smithers, DDS, Berkeley, California, United States of America


The unique capability of neutrons to penetrate metals and reveal internal structure of various objects enables non-destructive investigations in various fields, which do not include dental sciences at this time, except for the very few exclusions. Unfortunately, neutron radiography of teeth can be performed only ex vivo as neutrons are very harmful to humans. However, neutron radiography of extracted crowned teeth may provide complementary information to many other more conventional methods as neutrons enable non-destructive investigation of structures underneath metal crowns and behind metal fillings. Neutrons interact with some light materials, such as hydrogen, while their cross section for many metals is relatively low.

In this study we compare conventional X-ray imaging with neutron radiography and neutron tomography of several crowned teeth and teeth with amalgam fillings. While X-rays provide great images of the root area, they cannot probe anything in the area where metals are present, e.g. underneath the crown or behind the metal filling. We demonstrate here that neutron radiography and neutron tomographic images reveal the internal structure of treated teeth underneath the crown. This imaging method can be used to investigate the interaction of various fillings and crown attachment paste with the crown and the tooth itself. The integrity of tooth and fillings underneath the crown, the location of cavities and voids can be visualized in three dimensions and provide means for the improvements of filling and crown attachment materials and methods.

Keywords: Radiography, Dental imaging, Neutron radiography, Neutron tomography
Poster panel: 135

Poster Number:

Beta-ray imaging system with γ-ray coincidence (#2785)

T. Fukuchi1, S. Yamamoto2, J. Kataoka3, K. Kamada4, A. Yoshikawa4, Y. Watanabe1, S. Enomoto1

1 RIKEN, Center for Biosystems Dynamics Research, Kobe, Japan
2 Nagoya University, Graduate School of Medicine, Nagoya, Japan
3 Waseda University, Research Institute for Science and Engineering, Tokyo, Japan
4 Tohoku University, New Industry Creation Hatchery Center, Sendai, Japan


We have developed a β-ray imaging system, which can simultaneously detect β-ray and coincidence γ-ray. The system consists of a position sensitive β-ray detector and a large volume γ-ray detector. These detectors were mounted on the flexible frames and can locate on the arbitrary positional relations. A 40×40×1 mm3 La-GPS(Ce) scintillator was employed for the β-ray detection. This detector was pixelized in the pitch of 300 µm by a grooving process and coupled to a multi-anode position sensitive photo-multiplier tube for the position extraction. A bismuth germanium oxide (BGO) scintillator of a size of 50×43×16 mm3 coupled to a photomultiplier tube was used as a γ-ray detector. The system signal processor was designed to divide the data of β-ray detection into two groups in accordance with presence or absence of γ-ray coincidence. The energy window of the BGO detector is adjustable for the γ-ray energy. The different images can be reconstructed by the presence or absence of γ-ray coincidence. If β-γ emitters, which emit a prompt γ-ray successively after the β-ray, are used as the tracer, β-ray imaging with γ-ray coincidence has two advantages. One is a multiple-tracer imaging by identification of the radio-nuclide using the information of the prompt γ-ray, which has an intrinsic energy for the radio-nuclide. We succeeded in the multiple-tracer simultaneous imaging of 22Na (β-γ emitter) and 45Ca (pure β emitter). The other advantage is that γ-ray coincidence works as a γ-ray rejecter in the β-ray detection. Generally, a mean-free-path of γ-ray in the detector is longer than that of β-ray, and the long mean-free-path of γ-ray results in a blurred image. Therefore the β-ray imaging without a γ-ray mixing by the γ-ray rejection provides an improvement of the image quality. We confirmed this advantage by a phantom experiment.

Keywords: β-ray, γ-ray, imaging systm
Poster panel: 138

Poster Number:

An investigation of image quality for a mobile adaptive versus fixed detector geometries for awake animal SPECT (#2879)

D. Zhang1, G. I. Angelis2, J. E. Gillam3, S. R. Meikle2, T. Ma1, P. L. Kench2

1 Tsinghua University, Department of Engineering Physics, Beijing, China
2 University of Sydney, Brain and Mind Centre, Sydney, Australia
3 The National Centre of Excellence in Youth Mental Health, Orygen, Parkville, Australia


Single-Photon Emission Computed Tomography requires the animal be anaesthetized to prevent motion artefacts in reconstructed images. We are developing a purpose-built SPECT system for awake animal imaging using stationary detectors and a mobile (adaptive) detector geometry with integrated motion tracking and correction for a rigid structure. In this work, we investigate the ideal stationary detector geometry and characterize the impact of the tracking mobile adaptive detector on the image quality of awake animal SPECT in simulation. The simulated awake mouse moves with 6-degrees-of-freedom within the housing. The system consists of stationary detectors with parallel-hole collimation below, with pinhole collimated detectors in front of and beside the housing. An additional mobile (6-degrees-of-freedom) detector with pinhole collimation continuously follows above the moving animal to keep the area of interest within the detector field of view. Different detector geometries were simulated: below; front and side; below, front and side; above, front and side; and above, below, front and side. We tested two ways of above detector movement: fixed above, or tracking the mouse adaptively. A motion-compensation OSEM algorithm was used to reconstruct the images. Image quality was assessed using metrics of noise, contrast, and correlation of the reconstructed volumes to the true simulated image. We have demonstrated, by simulation, that the proposed SPECT system geometry permits awake animal imaging of a rigid structure such as the head. Preliminary results indicate reconstructions with the adaptive detector offers benefits in terms of visual and image quality measures of noise, contrast recovery and correlation when compared to fixed geometry. In future work, we will characterize the impact of the adaptive nature of the mobile detector on the image quality of simulated awake animal SPECT on a range of different of animal motion patterns.

Keywords: SPECT, awake animal imaging
Poster panel: 141

Poster Number:

Image-guided system for tumor ablation with magnetic particles (#1487)

I. N. Weinberg1, L. O. Mair2, S. Jafari2, O. Hale1, J. Watson-Daniels1, J. Baker McKee1, P. Y. Stepanov1, S. T. Fricke3

1 Weinberg Medical Physics, North Bethesda, Maryland, United States of America
2 Neuroparticle Corporation, North Bethesda, Maryland, United States of America
3 Children's National Medical Center, Radiology, Washington, United States of America


Currently, magnetic particle heating has been shown to improve survival of patients with brain tumors. However, placement of the particles requires a separate injection of particles to each tumor, with no ability to conformally direct the particles to the tumor margins. Local ablation of metastases to liver has been shown to significantly improve patient survival. Traditionally, patients with more than three hepatic metastases, or with any tumor sites greater than 5 cm in diameter, have been excluded from such local therapy. To generalize favorable outcomes to more patients, it would be useful to be able to more readily apply ablation techniques to patients with larger site numbers and/or sizes of metastases (who currently might be considered unsuitable for surgery). 

An experimental image-guided therapy system has been developed to ablate multiple tumor sites using conformally-localized magnetic particles heated with alternating magnetic fields. The new proposed image-guided modality utilizes magnetic pulse sequences (capable of concentrating and heating ferromagnetic microparticles deep in the body) in a novel MRI design to achieve rapid and conformal tumor ablation. An arbitrary number of metastatic sites could be targeted after a single intra- injection of particles, with subsequent real-time MRI-guided concentration of the particles to multiple locations with sub-millimeter precision. Once placed in the target locations, alternating magnetic fields would rapidly heat the particles to ablate targeted tumor sites. Preliminary data are presented demonstrating sub-millimeter resolution of heating that is fast enough to overcome physiological diffusion.

Keywords: image-guided therapy, mri
Poster panel: 144

Poster Number:

An approach for a reconstruction-derived whole-blood arterial input function (RDIF) in PET/MRI (#1843)

G. Schramm1, A. Rezaei1, M. Koole1, F. E. Boada2, K. van Laere1, J. Nuyts1

1 KU/UZ Leuven , Department of Imaging and Pathology, Division of Nuclear Medicine, Leuven, Belgium
2 New York University, Department of Radiology, NYU School of Medicine, New York City, New York, United States of America


An accurate arterial input function (AIF) is a crucial element required for
absolute quantification using tracer kinetic modeling in PET.
At the moment, continuous or manual sampling using arterial catherization
remains the gold standard to determine the AIF in brain PET studies.
Because arterial sampling is invasive and complicates the PET acquisition
workflow, several groups have investigated image-derived calculation of the
whole-blood AIF (IDIF) using a region of interest (ROI) in the carotid arteries.
Since the diameter of the carotid arteries is comparable to the achievable
reconstructed spatial resolution of current PET scanners (ca 5\,mm),
all IDIF methods are strongly affected by spill-out from the arteries to the background
and by spill-in from the background into the arteries.
Correction of the latter, requires precise knowledge of the local background
around the arteries which is in general non-trivial.

To overcome the latter limitation, we propose a method to directly reconstruct
the whole-blood AIF during iterative image reconstruction (RDIF).
In the RDIF approach, we aim to directly reconstruct the uptake of a fuzzy
arterial ROI taken from a segmentation of a
simultaneously acquired high resolution MR angiography image.

Reconstructions of simulated 3D time-of-flight PET data as well as reconstructions
of two human data sets acquired on a GE SIGNA PET/MR with measured whole-blood input function
revealed excellent agreement between the RDIF and the reference input function.
Moreover, we show that Total Variation regularization of the background accelerates
the convergence but also influence the bias of the reconstructed arterial uptake.

Keywords: PET/MR, quantification, arterial input function, kinetic modeling, iterative image reconstruction
Poster panel: 147

Poster Number:

NEMA-NU2 performance evaluation of a whole-body PET prototype with four-layer DOI detectors (#2188)

G. Akamatsu1, H. Tashima1, Y. Iwao1, H. Wakizaka1, T. Maeda1, A. Mohammadi1, S. Takyu1, F. Nishikido1, H. J. Rutherford2, A. Chacon2, M. Safavi-Naeini2, E. Yoshida1, T. Yamaya1

1 National Institute of Radiological Sciences (NIRS-QST), Chiba, Japan
2 Australian Nuclear Science and Technology Organisation, Sydney, Australia


One of the major issues in PET scanners is the parallax error caused by the thickness of the crystals. Spatial resolution at the peripheral regions of the field-of-view (FOV) is degraded. For this issue, depth-of-interaction (DOI) measurement is required to achieve higher and more uniform spatial resolution over the entire FOV. In this paper, therefore, we develop the world’s first whole-body PET prototype by using our original 4-layer DOI detectors. The performance measurement is carried out based on the NU2-2012 standard of the National Electrical Manufacturers Association (NEMA). In addition to the NEMA NU2 evaluation, we show the impact of DOI measurement by using the NEMA NU4 image quality phantom. As a result, the average spatial resolutions of 2.1±0.4 mm were achieved with the ordered-subset maximization-expectation (OSEM) reconstruction. The radial spatial resolution was improved with the DOI information by 48% at the 20 cm offset position. The contrast of small uptakes at the peripheral FOV was improved by the DOI information, which indicates the effectiveness of the whole-body DOI-PET for obese patients that occupy most of the FOV. Higher and more uniform spatial resolution and higher contrast for small lesions will enable improved detection of small or distal lesions such as nodal metastases.

Keywords: PET, DOI, Performance evaluation, NEMA
Poster panel: 150

Poster Number:

Improved Line-of-Response Allocation for Inter-Crystal Compton Events in PET (#2748)

C. Kim1, X. Jin2

1 GE Healthcare, Waukesha, Wisconsin, United States of America
2 GE Healthcare, Waukesha, Wisconsin, United States of America


In a PET detector, a gamma ray interacts through photoelectric absorption or Compton scattering. To locate the gamma ray direction for the line-of-response, the flood map of the detector was used. The flood map typically consists of the energy centroid of gamma ray events distributed in two-dimensional space. Based on the flood map, a crystal identification map was drawn and all gamma ray events around a crystal peak were allocated to the corresponding crystal. This has been done for both photoelectric and Compton events. However, this is mostly not correct for those inter-crystal Compton scatters. Since 511keV has higher probability in the forward scattering, it deposits more energy in the second interaction than the first. When the deposited energies were weighted to identify the direction of the gamma ray, the second interaction or second crystal gets more weight and so, assigned for the line-of-response, which is not correct. In short, the energy centroid method misallocates most of inter-crystal Compton scatters to wrong crystals. In this paper, we investigate the amount of misallocation on two different crystal geometries and propose an algorithm to improve its allocation.

Poster panel: 153

Poster Number:

PET Data Acquisition: Preparing PETLINK for the Coming Decades (#2497)

W. F. Jones1, A. P. Moor1, I. J. Huff1

1 Siemens Medical Solutions USA, Inc., MI / PLM-R&D RCC, Knoxville, Tennessee, United States of America


The Siemens PETLINK Guideline has helped make PET more effective in clinical data acquisitions for more than 20 years. Backward compatibility, reuse of functionality and publically shared formats common between product lines are part of these benefits. Of particular interest is the 64-bit detector-pair event (and tag) packet list-mode data format as defined. During those 20 years, the number of active bits used within such 64-bit packets increased from 30 bits in 2002 to 59 bits in 2018. So, our current 64-bit packet format is nearing capacity limit. We present our “works-in-progress” plans for expanding beyond these limits. Our first new format is a 96-bit packet (an extension of our current 64-bit format) which adds 31 available payload bits per packet; but many such bits will be unused leading to payload efficiency as low as 66%. Our second new format represents our “PETLINK-Frame” concept. Here we define a more complex, very flexible format, primarily supporting “bit packing” to significantly gain payload efficiency. In this case and for the most part, bit packing means the list-mode data stream is only burdened by the essential event packet bits – i.e. inactive event packet bits are absent. Such PETLINK Frames include a header/trailer to “frame” successive portions of the packed packet stream. The header supports synchronization with and parsing of the stream. The trailer provides error detection. With this concept, full flexibility is supported regarding the number of active payload bits selected for the bit-packed event packets held constant for a given stream – e.g. from 63 to 93 bits. In contrast with the 96-bit unpacked format example of 66%, payload efficiency for most any high-count-rate clinical stream using PETLINK Frames is shown to be 99.7% or better. We offer these concepts well before a critical need in PET development arises. In so doing, we hope to gain important, early feedback from the PET community and, thus, to continue our support of PET.

Keywords: PET Data Acquistion, PET List-Mode Data, Bit Packing, PETLINK
Poster panel: 156

Poster Number:

Design of a Cost-Effective High-Sensitivity Clinical PET/CT System with BGO and SiPM based Detectors: Preliminary Results and System Simulation (#2636)

X. Deng1, G. Fu2, H. Yan1, X. Ma1, J. Guo2

1 FMI Medical Systems, Inc., Solon, Ohio, United States of America
2 Eastern Blue Technologies, Inc, Ballston Spa, New York, United States of America


In clinical PET applications, the sensitivity is crucial for achieving high quality images. The sensitivity is mainly determined by the detection efficiency and the solid angle coverage of the scanner. In this work, we show the design of a cost-effective high sensitivity PET scanner by extending the axial FOV up to 40cm. The system sensitivity is expected to exceed at least two times of all current clinical PET scanners. BGO and SiPM PET detectors are chosen for their lower price as well as the SiPM’s superior quantum efficiency to conventional PMT. The PET/CT system integrates the PET scanner with a low-cost 16-slice CT.  The PET scanner consists of 2 to 8 detector rings axially for various sensitivity configurations. Each ring consists of 48 detector modules, with 2x2 blocks each. A block consists of a 6*6 array of 4.0*4.0*30 mm3 BGO crystals. A 1-to-1 coupling of crystal to SiPM is used, and a front-end board is designed to readout the energy and timing signals separately. The measured block metrics were applied in Monte Carlo simulations to estimate the system sensitivity and count rate performances for a 4-ring configuration (20cm axial FOV). An energy window of 425-650keV and a coincidence timing window of 9.5 ns were applied. The measured energy resolution was 15% at 511keV, and the coincidence resolving time was less than 5.5ns. The simulated sensitivity was 12.0kcps/MBq at center FOV. The peak NECR was 150kcps at 20.5kBq/ml, and the scatter fraction was 34.7%. The BGO-SiPM detector blocks were evaluated and the simulated system showed superior performance to other PET scanners. A 4-ring PET scanner is under construction, and comprehensive simulation is underway for all configurations.

Poster panel: 159

Poster Number:

Influence of image reconstruction parameters for target volume estimation with PET (#2725)

H. Gabrani-Juma1, D. La Russa2, R. Klein1

1 The Ottawa Hospital, Division of Nuclear Medicine, Ottawa, Ontario, Canada
2 The Ottawa Hospital , Department of Medical Physics, Ottawa, Ontario, Canada


For patients with non-small cell lung cancer (NSCLC), pre-treatment 18F-FDG-PET imaging is currently the standard of care in the staging of potentially curative disease. Information from 18F-FDG-PET scans is also used to decide on target volumes in those patients receiving radiation therapy and, increasingly, to define tumour sub-volumes to receive additional radiation dose. The latter strategy of boosting PET-avid regions to higher doses of radiation is motivated by recent evidence establishing a link between 18F-FDG-PET avidity and disease recurrence. In this work, we aim to determine the PET scan volume dependencies on image reconstruction algorithms and parameters, as well as segmentation techniques. The NEMA phantom was imaged on a Discovery 710 PET/CT containing a solution of 0.2 MBq/mL of FDG in each sphere, and 20 MBq of FDG injected as background. OSEM and BSREM reconstructions were used while image matrix size, beta (BSREM), number of iterations and subsets (OSEM) were varied. Delineation was estimated using relative threshold (RT) and maximum gradient (MG) techniques. ANOVA analyses were used to determine significance of contributing factors. Errors spanning between -65% to +590% were reported for the RT method while estimates for the MG method reported errors spanning from -44% to 34%. For OSEM images, interactions between the number of subsets and the number of iterations yielded significance (p < 0.01). Varying the image matrix size did not significantly affect estimates (p>0.05). In BSREM images beta and image matrix size both contributed to variability for volumetric measurements (p < 0.01). 

Keywords: PET, OSEM, BSREM, Q.Clear, Volumetric Measurements
Poster panel: 162

Poster Number:

Development of a MR Compatible Brain PET Insert with compact DOI-PET Detector and White Rabbit Data Synchronization Technique (#2798)

T. Xu1, Q. Wei2, H. Li3, G. Gong3, H. Gong3, Y. Liu3, T. Ma3

1 Beijing Novel Medical Equipment Ltd., Beijing, China
2 University of Science and Technology Beijing, Department of Automatic and Electrical Engineering, Beijing, China
3 Tsinghua University, Department of Engineering Physics, Beijing, China


A standardized DOI PET detector module has been developed based on staggered crystal block, SiPM array and self-developed 64-channel readout ASIC chips. In this work, such detector technology is applied to the development of a MR-compatible PET insert for brain imaging. The PET ring is composed of 15 singles detection units with a 330 mm bore size and 480 mm outer diameter. Each unit is organized in detector stacks containing 12 detector blocks arranged in 2×6, making the PET insert has 200 mm axial field of view. In DAQ electronics development, the open-source White Rabbit (WR) is used, which can provide sub-nanosecond accuracy and picoseconds precision of synchronization for distributed systems connected with Ethernet.  Preliminary MR compatibility tests were performed and initial results including crystal identification, energy resolution and timing resolution showed that minimal influence of MRI field on PET detector performance of PET detector. The integration of the whole PET system is under way.

Keywords: Brain PET/MR, DOI, SiPM, ASIC, White Rabbit
Poster panel: 165

Poster Number:

A complete scheme of empirical beam hardening correction using Grangeat consistency condition (#1267)

S. Abdurahman1, 2, R. Frysch1, 2, O. Beuing3, G. Rose1, 2

1 Otto von Guericke Universität Magdeburg, Institute for Medical Engineering, Magdeburg, Saxony-Anhalt, Germany
2 Otto von Guericke Universität Magdeburg, Research Campus STIMULATE, Magdeburg, Saxony-Anhalt, Germany
3 University Hospital Magdeburg, Institute for Neuroradiology, Magdeburg, Saxony-Anhalt, Germany


The polychromatic X-ray spectrum and the energy-dependent attenuation coefficients of the imaged object introduce beam hardening artifacts in CT reconstructed volumes. Cupping, streak, and spill-over are the different manifestations of beam hardening artifacts. Cupping or first order artifacts can be pre-corrected by the polynomial transformation of polychromatic projections. Polynomials are usually pre-computed using calibration with a homogeneous water phantom. However, the water correction alone is not adequate to correct the higher order artifacts due to beam hardening in bone. The residual spectral artifacts due to bone beam hardening can be reduced using iterative beam hardening correction methods. Here, the bone corrected projections are modeled using water pre-corrected projections, forward projections of the bone volume and the correction coefficients. In this paper, we present a calibration-free beam hardening correction using consistency conditions derived from Grangeat’s fundamental relation. Our multi-pass algorithm corrects first and higher order artifacts sequentially. During the first pass of the algorithm, the polynomial for water correction is estimated by enforcing the consistency condition on the uncorrected projection data. Subsequently, the prior volume is reconstructed from water corrected projections, and the bone structures are segmented using threshold segmentation. Forward projections of the bone volume are computed using Joseph forward projector. During the second pass of the algorithm, the correction coefficients necessary for bone beam hardening correction are estimated using the same consistency condition. Our results from the simulation and clinical datasets show that the global cupping and higher order artifacts are significantly reduced after the proposed correction. The algorithm does not require calibration or prior knowledge about the X-ray spectrum, material attenuation properties, and the detector response.

Keywords: Beam hardening artifacts, Consistency conditions
Poster panel: 168

Poster Number:

MARS Pulmonary Spectral Molecular Imaging: Potential for Locating Tuberculosis Involvement (#1501)

C. D. Lowe1, A. Ortega-Gil2

1 University of Otago, Christchurch, Department of Radiology, Christchurch, New Zealand
2 University of Madrid, Instituto de Investigación Sanitaria del Hospital Gregorio Marañón (IISGM), Madrid, Spain

On behalf of the MARS collaboration: Sikiru A Adebileje c, e, Maya R Amma c, Nigel Anderson c, Marzieh Anjomrouz a, Fatemeh Ashariomabad c, Ali Atharifard a, Benjamin Bamford c, Stephen T Bell a, Srinidhi Bheesette c, d, Anthony P H Butler a, b, c, d, e, Philip H Butler a, b, c, d, e, Alexander I Chernoglazov a, e, Tara Dalefield b, Niels J A de Ruiter a, b, c, e, Robert M N Doesburg a, Neryda Duncan b, Steven P Gieseg b, c, Brian P Goulter a, Sam Gurney c, Joseph L Healy a, b, Peter J Hilton b, c, Preveenkumar Kanithi b, e, Tracy Kirkbride f, Stuart P Lansley a, V B H Mandalika a, b, e, Emmanuel Marfo c, Mahdieh Moghiseh a, c, David Palmer g, Raj K Panta a, c, Hannah M Prebble a, b, Aamir Y Raja c, Mohsen Ramyar c, Peter Renaud b, c, Nanette Schleich h, Emily Searle b, Muhammad Shamshad a, Jereena S Sheeja c, Rayhan Uddin b, Lieza Vanden Broeke b, Vivek V S a, E Peter Walker c, Michael F Walsh a a MARS Bioimaging Limited, Christchurch, New Zealand b University of Canterbury, Christchurch, New Zealand c University of Otago Christchurch, Christchurch, New Zealand d European Organisation for Nuclear Research (CERN), Geneva, Switzerland e Human Interface Technology Laboratory New Zealand, University of Canterbury, Christchurch, New Zealand f Ara Institute of Canterbury, Christchurch, New Zealand g Lincoln University, Lincoln, New Zealand h University of Otago Wellington, Wellington, New Zealand


The aim of the present study is to show that non-invasive MARS photon-counting CT can differentiate between infected and healthy pulmonary tissue using iodine-based contrast agent at high resolution. Four C57BL/6J mice with chronic Tuberculosis (TB) were euthanized with CO2 and the pulmonary tissue excised. The TB lungs were incubated in 3% iodine solution. Mouse pulmonary tissue free of TB was also excised and incubated in the iodine solution for control purposes. Calibration of the MARS spectral CT involved scanning a phantom containing four concentrations of iodine along with water (soft tissue) and lipid (fat). The calibration phantom, control, and TB infected tissue were imaged at four threshold energy levels; 20, 27, 34, 45 keV, at a constant 60 kVp tube voltage and 90 μA tube current. Following analysis of the calibration phantom, material decomposition (MD) was applied to the pulmonary tissue samples and iodine to obtain material images. MARS Vision software was used to visualize the materials to produce 3D spectral images. TB granulomas are visible within the lung lobes due to the iodine uptake. Amount of iodine uptake can be measured in mg by analysis of the material images using MARS Vision. MARS spectral CT was able to better differentiate between infected and healthy tissue. The present study demonstrated non-invasive, photon-counting CT is capable of differentiating between infected and healthy tissue. Future studies will consider development of TB markers, or drug markers labelled with gold nanoparticles, to enhance the understanding of the basic biology and mechanisms underpinning TB, and its relevance to the phenomenon of persistence in the infected host during therapy.

Keywords: Spectral CT, Molecular Imaging
Poster panel: 171

Poster Number:

Non-invasive analysis of murine knee osteoarthritis model with grating-based X-ray phase-contrast tomography (#1709)

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

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


Osteoarthritis (OA), the most common type of arthritis, not only shows common affliction of aged people, but occurs even in young people as well. A precise and quantitative imaging of structures of bones and soft tissues simultaneously is of particular importance in the study, diagnosis and treatment of the gradual progression of OA. The recent development of X-ray phase contrast imaging with grating interferometry and conventional X-ray tubes, as a potential nondestructive imaging tool in biology and medical tests, which provides multiple information and higher soft tissue contrast comparing with conventional absorption-based X-ray radiography, makes a high resolution radiographic method for soft tissues and fine structures of bones and joints possible. In this study, an ex-vivo animal model of murine knee osteoarthritis was developed. A non-invasive (of the joint structure) analysis with grating-based X-ray phase contrast tomography was performed, whose results reveal ideal differentiation of different tissue, various traces of joint illness, advantage with current imaging methods as MRI, and show great potential in the future study of clinical diagnosis of OA. Due to the limitations of the submitted summary, further comparison and detailed analysis will be later presented.

Keywords: phase-contrast imaging, CT, knee osteoarthritis
Poster panel: 174

Poster Number:

Simulation study of acute subarachnoid hemorrhage using water density images of dual energy CT (#1779)

Y. Endo1, T. Koike1

1 Kyorin University, Graduate School of Health Sciences, Mitaka-shi Tokyo, Japan


Dual energy CT(DECT) in the data domain enables to generate the material density images(MDI). The purpose of this study was to investigate the possibility of detecting small amount of acute subarachnoid hemorrhage(SAH) diluted by cerebrospinal fluid(CSF) using water density images(WDI). We also evaluated the effect of separation between two energy spectra on detectability. Numerical simulation was performed using energy spectrum data and attenuation data of substance. Acute SAH was simulated by mixing of CSF and blood (hematocrit[Hct], 45%) ranging from 0% to 100%. For a test phantom with regions of different blood concentration, we generated 120kVp single energy CT images(SECT) and WDI. We generated WDI in three ways: spectra of x-ray tube at 80kVp and 120, 150, 150kVp with 0.5-mm tin filter addition(+Sn). In the four images, the mean value and the standard deviation in the regions of different blood concentration were calculated, and plots of the SAH/CSF ratio versus Hct were made. The effect of energy separation was evaluated by statistically significant difference between CSF and SAH in three WDI images. The SAH/CSF ratio increased linearly with all images as Hct increased, and WDI was higher than SECT in its slope. The standard deviation of WDI was decreased by separating the two energy spectra. In significant difference with CSF, the minimum blood concentration was 13.5% at 80-120kVp and 80-150kVp, and 9% at 80-150kVp(+Sn). In this study, it was found that the WDI using DECT improves the detection ability of smaller amount of acute SAH than SECT. In addition, it was suggested that more accurate detection was possible by separating the two energy spectra used for generating the MDI.

Keywords: Dual energy CT, Water density images, Subarachnoid hemorrhage, Energy spectrum
Poster panel: 177

Poster Number:

Non-uniformity Correction for Photon-counting Detectors Using Neural Network (#1844)

W. Fang1, L. Li1

1 Tsinghua University, Department of Engineering Physics, Beijing, China


Non-uniformity correction is a big challenge in practical use of photon-counting detectors for their spectral-dependent variation of thresholds and energy responses among different detector elements. If not corrected, this inhomogeneity sensitivity among detector elements will lead to strip artefacts in projection and ring artefacts in reconstructed images for photon-counting CT imaging. Many pre-calibration based frameworks and iterative methods have been proposed to solve this problem. But these methods either lack the generality for different imaging objects or suffer from high computational cost. Here we propose a novel deep learning based method for photon counting detectors non-uniformity correction. Residual network was used to estimate the format of non-uniformity artefacts of input images in three domains and artefacts-free images can be obtained by getting the estimated ring artefacts subtracted from original images. Quantized simulation results and experimental results show that our proposed network has a strong capability of estimating ring artefacts and we found that the model trained from simulation data set can estimate ring artefacts of real photon counting detectors well.

Keywords: spectral CT, photon-counting detector, non-uniformity correction, Neural Network, deep learning
Poster panel: 180

Poster Number:

First living human images from a MARS photon-counting 8-energy CT (#2026)

P. A. H. Butler1, 2

1 University of Otago, Radiology, Christchurch, New Zealand
2 University of Canterbury, School of Physical and Chemical Sciences, Christchurch, New Zealand

MARS Collaboration: Sikiru A Adebileje, Maya R Amma, Nigel Anderson, Marzieh Anjomrouz, Fatemeh Ashariomabad, Ali Atharifard, Benjamin Bamford, Stephen T Bell, Srinidhi Bheesette, Anthony P H Butler, Philip H Butler, Pierre Carbonez, Alexander I Chernoglazov, Tara Dalefield, Jerome Damet, Niels J A de Ruiter, Robert M N Doesburg, Neryda Duncan, Steven P Gieseg, Brian P Goulter, Sam Gurney, Joseph L Healy, Peter J Hilton, Preveenkumar Kanithi, Tracy Kirkbride, Stuart P Lansley, Chiara Lowe, V B H Mandalika, Emmanuel Marfo, Mahdieh Moghiseh, David Palmer, Raj K Panta, Hannah M Prebble, Aamir Y Raja, Mohsen Ramyar, Peter Renaud, David Rundle, Nanette Schleich, Emily Searle, Muhammad Shamshad, Jereena S Sheeja, Rayhan Uddin, Lieza Vanden Broeke, Vivek V S, E Peter Walker, Michael F Walsh


The first diagnostic quality images of living humans have been obtained with a large bore MARS photon-counting spectral (8-energy) CT scanner. This result enables human clinical research using MARS technologies. Pre-clinical users of MARS scanners have had significant results in the fields of cancer, pharmacology, bone health, and vascular disease. This pre-clinical work was done with cell cultures, ex-vivo specimens, and murine models. With the large bore scanner it can now be translated to research with humans, sheep, or pigs. The MARS large bore scanner rotates an x-ray camera and x-ray tube around the object in spiral geometry. The 700mm bore enables a wide range of objects to be imaged. The system is highly flexible with variable magnification, tracking collimators, multiple x-ray filters, and programmable scan trajectory. A standard 120kVp x-ray tube enables penetration of human-sized objects. The x-ray camera uses multiple Medipix3RX detectors bonded to 2mm CZT. These operate in spectral mode with a pixel pitch of 110 μm. The x-ray camera contains at least 5 detector modules, but with flexibility to extend this to an array of up to 700 mm x 28 mm. Medipix3RX records 8 energies simultaneously using a variety of modes including some with charge sharing suppression schemes (known as Charge Summing Mode). The camera can be translated to enable imaging of objects where the field of view is greater than the detector array. Projection data is reconstructed into 3D volumes of x-ray attenuation data using an in-house algebraic technique. This is able to reconstruct from arbitrary geometries. Once 3D x-ray attenuation is obtained a second step is used to convert attenuation data to 3D volume of partial density of basis materials (mg/cm3). An interactive 3D viewing program has been developed to visualise and quantify the materials within the scanned object.

Keywords: Spectral CT, Photon Counting, Radiology, human imaging, computed tomography
Poster panel: 183

Poster Number:

Accurate Geometry Calibration Method in the Offset Cone-beam CT system (#2191)

J. Hwang1

1 KAIST, Nuclear and Quantum Engineering, daejeon, Republic of Korea


Geometry calibration is important in securing high
spatial resolution in cone-beam CT. A ball array phantom is
widely utilized for such a calibration. The algorithmic extraction
of geometry parameters from the projection of such phantom is
well established. However, in practical applications, such methods
may encounter challenging problems that may lead to suboptimal
calibration. In this paper, we explain two main causes of errors
that often occur in a geometry calibration of an offset cone-beam
CT system, and provide a remedy that can reduce such errors. The
validation of the proposed method was conducted using a real
offset cone-beam CT system.

Keywords: Geometry Calibration, Offset cone-beam CT system, Projection matrix
Poster panel: 186

Poster Number:

Analysis of the effects of silica coating of gold nanoparticles on dose enhancement and particle quantification using x-ray photon counting multispectral imaging (x-CSI) and photoacoustic spectral imaging (PASI). (#2415)

O. Pickford Scienti1, 2, A. Shah1, J. Bamber1, 2, D. Darambara1, 2

1 Institute of Cancer Research, Radiotherapy and Imaging, London, United Kingdom
2 Royal Marsden NHS Trust, Multi-modality Molecular Imaging team, London, United Kingdom


The ratio of radiotherapy dose delivered to the tumour and that delivered to healthy tissue is referred to as the therapeutic index (TI). A proposed method to increase the TI is the use of targeted dose enhancing agents such as gold nanoparticles (AuNPs). The clinical use of such agents requires a way of confirming their distribution in vivo for use in predicting the spatial distribution and intensity of the dose enhancing effects (DEE) that will occur. Silica-coated gold nanorods have been shown to improve PASI detectability as well as provide a range of clinically relevant benefits including reduced toxicity, improved cellular uptake and preservation of the photoacoustic spectral properties of the nanorods upon aggregation. This work aims to assess the effects of silica coating on DEE and on the accuracy of AuNP quantification using two techniques: x-ray photon counting multispectral imaging (x-CSI) and photoacoustic spectral imaging (PASI). The x-CSI work has involved a custom simulation framework (COGI), combining finite element analysis and Monte Carlo methods. A physical photon counting CdTe detector has also been used, with initial experimental measurements showing reasonable agreement between simulation and experiment, and an energy resolution of 4% at 136 keV. Calibration curves for AuNP concentrations are obtained and used to determine independent test concentrations of AuNPs. The PASI work used a preclinical MSOT® system. Results show that the silica coating largely prevents the signal degradation seen with naked AuNPS upon aggregation. Cell work to look at the effect of silica coating on DEE is also underway, with dose-response curves for a prostate cancer cell line (PC-3) obtained indicating that a dose of 4 - 6 Gy should maximise sensitivity to detect whether the silica coating alters the DEE.

Keywords: Photon counting, Spectral, Dose Enhancement, Gold Nanoparticles, Photoacoustic
Poster panel: 189

Poster Number:

Scatter Noise Rejection with Time-of-Flight Computed Tomography (#2610)

J. Rossignol1, Y. Bérubé-Lauzière2, R. Fontaine1

1 Université de Sherbrooke, Interdisciplinary institute of technological innovation 3IT, Sherbrooke, Québec, Canada
2 Université de Sherbrooke, Department of Electrical and Computer Engineering , Sherbrooke, Québec, Canada


Scatter noise in cone-beam computed tomography (CT) reduces the contrast-to-noise ratio (CNR), increases inaccuracy and causes cup and streak artifacts. This noise can be mitigated with an undesirable increase in radiation dose. We propose to use the time-of-flight (TOF) of X-ray photons to reduce the scattering noise, improve CNR without any increase of the radiation dose. Herein, a complete flat panel cone-beam scanner is simulated using GATE to assess feasibility and performance of time-of-flight computed tomography (TOF-CT) for this specific application. The simulated TOF CT scanner comprises a monochromatic X-ray source, irradiating with a cone angle of 16°, plexiglass phantoms of 50, 100, 150 and 200 mm thickness used to generate increasing levels of scatter radiation and a flat panel made of a 256 x 256 array of 1 x 1 x 1 mm3 detectors. A digital scatter rejection algorithm (DSR) compares the TOF of each individual photon with the expected TOF for a direct trajectory and removes the photons who failed to arrive in time. The DSR performance is evaluated first with single projection acquisitions for different configurations, then a full 360° acquisition is realised to measure the effect of DSR on image quality.  In a TOF-CT model with perfect time resolution and 120 keV photons, the scatter-to-primary ratio (SPR) can be reduced to less than 3%, even with an initial SPR as high as 300%. With 10 and 100 ps FWHM total resolution, SPR is reduced from 225% to respectively 8% and 40%. A full 360° acquisition of a water cylinder with two cylindrical bone inserts is simulated with a perfect time resolution. In this case, 100 keV photons are emitted in a 28° cone towards the detection system comprising a 256 x 256 array of 2 x 2 mm2 pixels. SPR decreases from 300% to 4% and CNR increases by 145% and the artifacts are greatly reduced with DSR. At a dose four times smaller, DSR obtained a CNR similar to the uncorrected image with a full dose.

Keywords: Scatter noise, Cone-beam Computed Tomography, Time-of-flight, Scatter rejection, Pulsed X-ray source
Poster panel: 192

Poster Number:

Research of Photon Counting Dual-energy X-ray Absorptiometry Based on SiPM (#2773)

Z. Wu1, Q. Du1, J. Mo1, B. M. W. Tsui2, M. Sun1, D. Xi3, G. Yuan1, J. Zheng1, X. Yang1

1 Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou, China
2 Johns Hopkins University, Department of Radiology, Baltimore, Maryland, United States of America
3 Raymeasure Medical Technology Co. Ltd, Suzhou, China


Recently there has be an increase of osteoporosis cases due to the increased of the ageing population. As an established clinical diagnostic tool for osteoporosis, dual-energy x-ray absorptiometry (DEXA) is widely used due to its accuracy, precision and image quality. In recent years, a new type of silicon based photomultipliers, SiPMs, has been developed. Its compact size and reliable performance is a potential photon counting DEXA detector. In this work we designed a DEXA test platform and tested the bone density imaging parameters of a SIPM detector. We constructed an experimental imaging platform with a fan-beam x-ray source from Spellman and the self-developed scanning motion control system. We studied the performance characteristics of a 64 channel YSO/SiPM array detector and compared them with a CZT detector. Moreover, the maximum-likelihood (ML) estimation method is used to optimize surface fitting algorithm and get higher material decomposition accuracy. The result show that the average decomposition error of CZT is 0.83% for PMMA and 0.76% for Al. The average error of the SiPM detector is 2% for PMMA and 2.74% for Al. Future studies will include the calibrations of the SIPM detector to increase the material decomposition accuracy. We conclude YSO/SiPM detector has the potential in DEXA imaging with good imaging performance characteristics at low cost.

Keywords: DEXA, SiPM, Bone Mineral Density
Poster panel: 195

Poster Number:

Cardiac and Respiratory Motion-Compensated Simultaneous Whole-heart PET-MR Imaging (#1264)

C. Munoz1, R. Neji1, 3, K. P. Kunze2, S. G. Nekolla2, R. M. Botnar1, C. Prieto1

1 King's College London, Department of Biomedical Engineering, London, United Kingdom
2 Technische Universität München, Nuklearmedizinische Klinik und Poliklinik, Munich, Germany
3 Siemens Healthcare, MR Research Collaborations, Frimley, United Kingdom


Image degradation due to cardiac and respiratory motion remains a challenge for cardiac PET-MR imaging. Although novel approaches for PET-MR motion compensation have been recently proposed, these schemes usually just acquire motion information from MR data during the PET acquisition and diagnostic MR afterwards, significantly increasing total exam time and leading to misaligned images. Here we propose a simultaneous dual-phase coronary MR angiography (CMRA) and cardiac PET acquisition and reconstruction framework that allows for simultaneous visualisation of coronary anatomy and motion-corrected myocardial PET in a single efficient examination. Non-rigid respiratory and cardiac motion is estimated from MR images, and the MR-derived deformation fields are used to correct both the CMRA data (respiratory motion correction for each cardiac phase) and the simultaneously acquired PET data (respiratory and cardiac motion correction). We tested this approach in healthy subjects and patients with cardiovascular disease, and observed that motion correction improved quality in CMRA images allowing for good depiction of the coronary arteries compared to a non-motion-corrected approach. Furthermore, results from patients with cardiovascular disease showed improvements in PET image quality when applying both cardiac and respiratory motion correction, resulting in sharper myocardial edges, and allowing the observation of small features such as the papillary muscles. Overall, the proposed approach produced high quality images in both modalities, demonstrating the potential for allowing a comprehensive non-invasive assessment of heart disease in a single and time-efficient examination.

Keywords: cardiac PET-MR, motion correction
Poster panel: 198

Poster Number:

Performance evaluation of an advanced detector module for an RF-penetrable TOF-PET insert for simultaneous PET/MRI (#1942)

Q. Dong1, I. Sacco1, C. - M. Chang1, 2, B. J. Lee1, 3, C. S. Levin1, 6

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


Simultaneous PET/MRI, which combines molecular and anatomical assessment, is a promising multi-modality imaging method. However, the high cost of the current commercial integrated PET/MRI systems limit the long-term potential of PET/MRI imaging. We are studying designs of a PET ring that can be inserted into existing MRI systems in order to reduce the cost without affecting the performance of either the PET or MRI components by leveraging the existing MR infrastructure. We are developing a second-generation RF-penetrable brain-dedicated TOF-PET insert with a trans-axial and axial FoVs of 28 cm and 16 cm, respectively. The PET ring comprises 16 detector modules, each hosting 6 sub-modules. Each sub-module comprises an array of 8 x 16 LYSO crystal elements (3.2 x 3.2 x 20 mm3) one-to-one coupled to an array of SiPMs read out by 4 36-channel readout ASICs. In this work, we evaluated the energy and coincidence timing resolution performance of a detector module, which was thermally regulated by a liquid chiller system at 24 degrees Celsius during the experiments. The energy resolution at 511 keV measured for 128 channels of one sub-module is 11.4 ± 0.9% (mean ± standard deviation) and preliminary coincidence timing resolution measured over two sub-modules in coincidence is 379.1 ± 25.7 ps FWHM. Potential factors that degrade the coincidence timing resolution are discussed.

Keywords: RF-penetrable, PET insert, TOF, PET/MRI
Poster panel: 201

Poster Number:

A fiber-based Laparo-PET/NIRF/visible multimodal system for sentinel lymph node and tumor detection for laparoscopic surgery (#2167)

S. H. Song1, Y. B. Han1, H. - Y. Lee2, D. H. Jeong3, H. G. Kang4, 1, S. J. Hong1, 5

1 Eulji University, Department of Senior Healthcare, Daejon, Republic of Korea
2 Seoul National Univsersity, Department of Nuclear Science Medicine, Seoul, Republic of Korea
3 Seoul National University, Department of Chemistry Education, Seoul, Republic of Korea
4 National Institute of Radiological Sciences, Department of Radiation Measurement and Dose Assessment, Chiba, Japan
5 Eulji University, Department of Radiological Science, Seongnam, Republic of Korea


Minimally invasive robotic surgery (MIRS) has been widely used because of its advantage of less incision, fast recovery time and efficiency. Conventional laparoscopic surgery using visible light and near-infrared fluorescence (NIRF) has a penetration depth problem (mostly < 10 mm). The purpose of research is to develop a multimodal imaging system showing an integrated 511 keV gamma-rays, NIRF, and visible image.
Previous research showed a feasibility of this multimodal system using optical fiber bundle and dichroic mirror module. However, severe light loss was occurred during the light transmission inside a fiber bundle. To overcome this drawback, electronic parts and other structures of the coincidence Laparo-PET detector were changed suitable for laparoscopic surgery (diameter of prior module = 25 mm, now < 15 mm possible and energy resolution below 25 %).
Images of 511 keV gamma, NIRF, and visible were simultaneously obtained and merged by improving not only individual parts and assembly of internal detector but also expansion of external detector (2 by 2 modules).

Poster panel: 204

Poster Number:

System design and detector performance of the PET component of the TRIMAGE PET/MR/EEG scanner (#2305)

N. Belcari1, 2, M. G. Bisogni1, 2, P. Carra1, 2, T. Cortopassi1, 2, N. Camarlinghi1, 2, M. Morrocchi2, G. Sportelli1, 2, A. Del Guerra1, 2

1 University of Pisa, Department of Physics "E. Fermi", Pisa, Italy
2 INFN - Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, Pisa, Italy


We present the final design and detector performance of the PET component of the TRIMAGE brain scanner. The full PET ring comprises 18 sectors. Its detectors are based on staggered LYSO crystals, 4 × 8 monolithic arrays of SiPMs and dedicated ASICs. The staggered crystals are made of 8 × 8 + 7 × 7 pixels. Each sector is 55 mm (transversal) × 163 mm (axial) size, each one consisting of three square detector modules hosted in an RF shielded cassette. Results show that the optimal performance is obtained at -29.5V bias voltage resulting in an excellent pixel identification and an energy resolution of 18% and 20% for the bottom and top layer, respectively. The average CTR measured for a pair of detector cassette is 650 ps.

Keywords: PET instrumentation, Brain PET, SiPM, performance evaluation
Poster panel: 207

Poster Number:

Coincidence processing and timing performance of the TRIMAGE PET system (#2360)

G. Sportelli1, 2, N. Belcari1, 2, M. G. Bisogni1, 2, N. Camarlinghi1, 2, P. Carra1, 2, M. Morrocchi2, M. Piendibene1, 2, A. Del Guerra1, 2

1 University of Pisa, Department of Physics "E. Fermi", Pisa, Italy
2 Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, Pisa, Italy


We present the coincidence processing and timing performance of the TRIMAGE brain PET system. The full PET ring comprises 18 detector cassettes. The detectors are based on staggered LYSO crystals, 4 × 8 monolithic arrays of Advansid NUV SiPMs and 64-channel TRIROC ASICs. The staggered crystals are made of 8 × 8 + 7 × 7 pixels. Each full detector consists of 12 crystal matrices, 24 SiPM arrays and 12 ASICs. The resulting sensitive area of each detector is 54 mm × 162 mm, the latter corresponding to the axial length of the field of view. Timestamp generation is performed by evaluating the pattern of energy deposition in the whole crystal matrix and correcting for the walk effect after pixel identification.
A custom coincidence processor has been also implemented and validated. The coincidence processor sorts the events incoming from the PET detectors and discriminates prompt coincidences in real time.
Results show that the best achieved CTR is 677 ps on the top layer and 631 ps on the bottom layer with the weighted t-fine method. Fine time correction accounted for up to 40% of the total CTR. Walk correction allows reducing the CTR by 9%.

Keywords: PET instrumentation, SiPM, time to digital conversion, CTR
Poster panel: 210

Poster Number:

Total Breast Imaging: Paradigm Shifting Concept for MR-Compatible Dual-Whole-Breast PET (#2753)

M. P. Tornai1, Y. - C. Tai2, S. Majewski3, M. B. Williams3, T. G. Turkington1, A. Z. Register1, S. Samanta2, J. Jiang2, J. A. O'Sullivan2

1 Duke University Medical Center, Radiology, BME, and Medical Physics, Durham, North Carolina, United States of America
2 Washington University, Radiology, and Electrical & Systems Engineering, St. Louis, Missouri, United States of America
3 University of Virginia, Radiology, BME, and Physics, Charlottesville, Virginia, United States of America


We present a new concept for Total Breast Imaging (TBI) by addressing many of the limitations of previous dedicated PET, PET/CT, and MR-compatible breast PET designs. This concept was collaboratively developed by our multi-institutional team. Our design criteria for MR-compatible Total Breast Imaging include the following: (1) both breasts can be imaged by PET, and ultimately, simultaneously with MR; (2) dynamic imaging in both breasts and chest wall volume is feasible; (3) the chest wall and axillary regions can easily and reproducibly be imaged; (4) images from both modalities are acquired in identical conditions and frames, and are easily registered; (5) the system has high geometric sensitivity and FOV coverage to exploit either lower dose PET imaging, or higher signal to noise ratio (SNR) and resolution imaging through lower image noise; (6) removable lateral detectors promote easy biopsy access in the MR reference frame. The unique racetrack design surrounding both breasts is augmented by anterior and posterior detector panels to enhance sensitivity and facilitate chest wall and axillary imaging. The scintillator-photodetector units incorporate excellent TOF resolution and DOI sensitivity, in order to better handle the oblique rays given the torso orientation. Data will be collected in list mode with post-acquisition coincidence processing. The fully-encompassing TBI geometry and anticipated capabilities are facilitated by improvements in solid-state, MR-compatible photodetector technology and electronics, as well as new scintillator materials. Initial GATE simulations show full volume imaging in the breast-related regions of interest, as well as nearly 7% sensitivity. The unique design is integrated into an MR-breast coil. In the research and clinical settings, the TBI system is meant to enable low dose and/or high SNR and resolution breast imaging in order to increase the lesion specificity of the already high-sensitivity MR imaging environment.

Keywords: PET, MR, Breast Imaging, PET/MR, Clinical
Poster panel: 213

Poster Number:

Simultaneous Multi-Dataset PET Image Reconstruction Using a Progressive Mutually Weighted Quadratic Penalty (#1101)

S. Ellis1, A. J. Reader1

1 King's College London, School of Biomedical Engineering and Imaging Sciences, London, United Kingdom


There are many contexts in which a patient is scanned with positron emission tomography (PET) multiple times, for example in oncology treatment monitoring scans, or to obtain complementary information from multiple radiotracers. As is common in PET, such scans often suffer from high noise due to the counts-limited nature of PET acquisitions. Reduction of this noise while preserving scan-specific features by sharing information between datasets is the subject of ongoing research. In this work we adapt a recently proposed joint PET-MR reconstruction method to the multi-dataset PET context in order to reduce PET image noise across data-series. The method calculates similarity weights between neighbouring voxels in each image at each iteration and uses these to update the weighted quadratic penalty (WQP) for use in the next iteration. By iterating the two-step process of weights calculation and a penalised PET reconstruction update, the mutually weighted quadratic penalty (MWQP) method is defined. The proposed method was applied to a 2-scan, 2D, [18F]fluorodeoxyglucose brain cancer treatment response simulation study. Results indicate that using the MWQP method allows for reduced reconstructed image noise (up to 89%) while avoiding bias within scan-unique features, such as tumours. In this respect the method is superior to maximum-likelihood expectation maximisation reconstructions with a Gaussian smooth, total-variation regularised reconstructions, and traditional guided image reconstruction techniques that only allow a one-way transfer of information between datasets. Reconstructed visual image quality reflects the quantitative results, demonstrating lower image-wide noise and higher tumour contrast. These results suggest that using a MWQP reconstruction approach improves reconstructed image quality beyond that achieved with conventional methods. Future work will involve further characterisation of the method and testing on clinical datasets.

Keywords: simultaneous image reconstruction, longitudinal PET
Poster panel: 216

Poster Number:

Total variation regularization for list-mode MLEM reconstruction in Compton camera imaging (#1238)

Y. Feng1, A. Etxebeste2, J. M. Létang1, D. Sarrut3, V. Maxim1

1 INSA Lyon, CREATIS, Villeurbanne, France
2 Université de Lyon, CREATIS, Villeurbanne, France
3 CNRS, CREATIS, Villeurbanne, France


The Compton camera is a gamma ray imaging device already employed in astronomy and still in investigation for clinical domain. A key point in the imaging process is the tomographic reconstruction step. When the acquisition parameters and the a priori information are correctly accounted for, iterative algorithms are able to produce accurate images by compensating for measurement uncertainties and statistical noise. In this work we focus on the list-mode maximum likelihood expectation maximization (LM-MLEM) algorithm with smoothness a priori information expressed by the total variation norm. This type of regularization is particularly well suited for low-dose acquisitions, as it is the case in the applications foreseen for the camera. We show that the TV a priori strongly improves the images when data are acquired in ideal conditions. For realistic data, this a priori is not sufficient and deconvolution with pre-calculated image-space point spread function should also be considered.

Keywords: Compton camera imaging, MLEM-TV
Poster panel: 219

Poster Number:

Low-dose Digital Tomosynthesis Reconstruction Using a Stationary Multislit Collimator (#1336)

S. Park1, G. Kim1, C. Park1, D. Lee1, H. Cho1, C. Seo1

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


An effective method to reduce radiation dosage in digital tomosynthesis (DTS) was investigated where a fixed multislit collimator was placed between the x-ray tube and the patient, partially blocking the x-ray beam to the patient thereby reducing the radiation dosage. In such a method, a sinogram measured is possibly interpolated using various mathematical functions to fill the missing data caused by the blocking collimator, which thereby alleviates bright-band artifacts in the reconstructed image. Thus, elaborate sinogram interpolation methods have a significant impact on the resultant image quality. In this study, a prior sinogram was first generated by selecting the area surrounding the primary harmonic peak in the frequency domain after Fourier transforming the measured sinogram because the multislit collimator acts as a frequency modulator for the original sinogram’s Fourier contents. The missing data of the measured sinogram were then filled out by a smoothing interpolation function with the prior sinogram. To validate the proposed method, we performed a simulation using a multislit collimator with 50% duty cycle and investigated the image characteristics. We reconstructed DTS images with no bright-band artifacts around the multislit’s edges using the filtered-backprojection algorithm, keeping the image quality not degraded visually. More details of the simulation results will be described in the paper.

Keywords: Low-dose, Digital tomosynthesis
Poster panel: 222

Poster Number:

A New ToF-PET Likelihood Model: Parameterizing the Angular Distribution of Emission (#1422)

M. Toussaint1, 2, J. - P. Dussault1, R. Lecomte2

1 Université de Sherbrooke, Computer Science, Sherbrooke, Québec, Canada
2 Université de Sherbrooke, Sherbrooke Molecular Imaging Center, CRCHUS, and Department of Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada


The advantages of using Time-of-Flight (ToF) to improve image contrast in Positron Emission Tomography (PET) has been studied extensively. Yet, its potential influence on spatial resolution was neglected up to now since the ToF resolution is still much larger than the physical resolution of PET scanners. With the expected progress in coincidence time resolution, we have shown previously that ToF information can be used to mitigate the blur induced by the detector width. In such cases, the ToF information must be finely discretized, which results in low statistics histograms that negatively affect the quality of images reconstructed using the likelihood model. In an attempt to circumvent this limitation, we propose a new model that explicitly parameterize the angular distribution of emission for each pixel. The resulting model has significantly more free variables than the classical approach and a penalization function enforcing the isotropic distribution of emission is implemented to better take advantage of the increased degrees of freedom. In order to validate the model, a Hot Spot phantom acquisition was simulated from which two ToF histograms were produced: one with coarse sampling introducing a loss of ToF information, and the other with adequate ToF sampling. Both of them were reconstructed using the proposed model and the ToF likelihood model and their respective optimal reconstructions were compared. The proposed model was able to reconstruct similar resolution as the classical model, but with better noise property. These preliminary results indicate that the proposed model has the potential to circumvent the limitation induced by low statistics histograms. A complete validation test suite of the strengths and limitations of the proposed model will be presented.


Keywords: PET Image Reconstruction, Modelization, Time-of-Flight, Iterative Algorithm
Poster panel: 225

Poster Number:

Preliminary investigation of a Monte Carlo-based system matrix approach for quantitative clinical brain 123I SPECT imaging (#1597)

B. Auer1, N. Zeraatkar1, S. Banerjee1, J. C. Goding1, G. I. Zubal2, L. R. Furenlid3, M. A. King1

1 University of Massachusetts Medical School, Dept. of Radiology, Worcester, Massachusetts, United States of America
2 Z-Concepts, LLC, East Haven, Connecticut, United States of America
3 University of Arizona, Dept. of Radiology & The College of Optical Sciences, Tucson, Arizona, United States of America


A next-generation, adaptive, dynamic multi-pinhole system, AdaptiSPECT-C, dedicated to clinical brain SPECT imaging, is currently under development as part of a collaboration between the universities of Arizona and Massachusetts. It has been shown that accurate modeling of the system matrix is a key aspect of SPECT image reconstruction as it has the potential to improve the imaging performance of any system. A straight-forward approach to modeling is based on the use of Monte Carlo simulation to pre-compute and store the system matrix. Generally, in clinical imaging, given the large sizes of detectors and volume of interests this approach faces critical memory storage issues despite the use of sparse structures to store the system matrix.
The aim of this work was to first investigate the feasibility of a Monte Carlo simulation pre-computed system matrix approach for 123I clinical brain SPECT imaging with the AdaptiSPECT-C system. Secondly, an efficient method based on a PSF spatial thresholding for reducing the system matrix size was evaluated using an XCAT brain perfusion phantom.
The present approach’s feasibility was fully demonstrated in case of clinical 123I brain imaging. Our method exhibited superior performance in terms of NMSE (0.14) and bias (0.1%), leading to a quantitative image reconstruction. Moreover, the associated reconstruction speed was 8 sec per iteration for our 23 detector design. The use of our approach could thus potentially facilitate the dynamic and adaptive imaging capabilities of AdaptiSPECT-C.
A matrix spatial thresholding method was successfully developed and validated against the case for which no threshold on PSFs was applied. Decreasing the number of matrix elements associated with low probabilities which are mostly responsible for noise propagation in reconstruction, significantly improved the results while reducing the matrix size.

Keywords: Clinical 123I brain imaging, modeling of the system matrix, Monte Carlo simulation, quantitative SPECT imaging, PSF spatial thresholding
Poster panel: 228

Poster Number:

Track-based multiple scattering tomography -- Simulation and in-beam measurements (#1671)

H. Jansen1, P. Schütze1

1 Deutsches Elektronen-Synchrotron (DESY), Hamburg, Hamburg, Germany


Tomographic methods for the imaging and visualization of complex structures are widely used in medical, industrial and scientific applications.
Computed Tomography with X-rays exploits the difference of attenuation length for photons in different materials. Complete absorption of the photon beam for large or highly absorbing materials poses a limit on the acceptable sample's material budget.

We propose a new imaging method based on the tracking of charged particles in the GeV range traversing a sample under investigation. By measuring deflection angles at the sample originating from multiple Coulomb scattering, a position resolved estimate on the projected material budget is extracted. This allows for the 3D-reconstruction of the material-budget making use of an inverse Radon transform. The large penetration range of GeV electrons or protons allows to study of samples with large material budget.

A phantom was studied both in simulation and in-beam using the AllPix Detector Simulation Framework and the DESY Test Beam Facilities. This allows for a detailed comparison between simulation and experiment. The experiment was carried out using a GeV electron beam and the DATURA Beam Telescope for high-precision particle tracking.

We present a detailed study on track-based multiple scattering tomography comparing simulation with in-beam results. The potential and limits of this tomographic technique are discussed in terms of acceptable material budget, spatial resolution and contrast-to-noise ratio. Additionally, a comparison with conventional CT is given exemplifying once more the potential of track-based multiple scattering tomography for samples of large material budgets.

Keywords: Charged Particle Tracking, Multiple Coulomb Scattering, Material Budget Imaging, Computed Tomography, Beam Telescopes
Poster panel: 231

Poster Number:

Filtered Backprojection Implementation of the Immediately-After-Backprojection Filtering (#1447)

G. L. Zeng1, 2

1 Weber State University, Egineering, Ogden, Utah, United States of America
2 University of Utah, Radiology and Imaging Sciences, Salt Lake City, Utah, United States of America


In an iterative image reconstruction algorithm, it was demonstrated that the contrast-to-noise ratio in the final reconstruction could be improved if a low-pass filter was applied to the backprojction of the projection-domain discrepancy, and then this backprojection was used to update the image from the previous iteration. The goal of this paper is to extend this method to the weighted filtered backprojection (FBP) algorithm.

Poster panel: 234

Poster Number:

Revisit of Unmatched Projector/Backprojector in an Iterative Algorithm with counterexamples (#1449)

G. L. Zeng1, 2

1 Weber State University, Engineering, Ogden, Utah, United States of America
2 University of Utah, Radiology and Imaging Sciences, Salt Lake City, Utah, United States of America


 It is rather controversial whether it is justified to use an unmatched projector/backprojector pair in an iterative image reconstruction algorithm. One common concern of using an unmatched projector/backprojector pair is that the optimal solution cannot be reached. This concern is a misleading and must be clarified. We define a figure-of-merit in the image domain as the distance between the reconstructed image and the true image, as the normalized mean-squared-error (NMSE). The NMSE is used to determine whether an unmatched matched projector/backprojector pair can provide a better image than a matched projector/backprojector pair. Hot and cold lesion’s contrast-to-noise ratio is also used as an alternative secondary figure-of-merit for algorithm comparison. Computer generated counterexamples are used to test the performance for matched and unmatched projection/backprojection pairs for different reconstruction algorithms. The projectors are ray-driven, and the backprojectors are ray-driven and pixel-driven. For the attenuation-free data examples, the unmatched pixel-driven backprojector outperforms the matched ray-driven backprojector. For the attenuated data example, the matched ray-driven backprojector performs better. The ray-driven backprojector can be slightly improved by using an attenuation coefficient that is larger than the true one; in this case the backprojector becomes unmatched. Conclusions: Unmatched projector/backprojector pairs have much flexibility. If the backprojector is properly chosen, good results can be obtained. However, we have not found a general rule to select a good backprojector.

Poster panel: 237

Poster Number:

Estimation of the Optimal Iteration Number for Minimal Image Discrepancy (#1450)

G. L. Zeng1, 2

1 Weber State University, Engineering, Ogden, Utah, United States of America
2 University of Utah, Radiology and Imaging Sciences, Salt Lake City, Utah, United States of America


Due to noise, the iterative image reconstruction algorithms must stop early before reaching the convergence. There is an optimal stopping point, at which the discrepancy of the reconstruction to the true image reaches minimum. However, it is still an open problem to find this optimal stopping point. This paper establishes two relationships towards solving this open problem. The first relationship is between the iterative Landweber algorithm and a weighted filtered backprojection (FBP) algorithm. The second relationship is between the optimal weighted FBP reconstruction and the optimal projection-domain filtered data. These two relationships can lead us to the optimal stopping point.

Poster panel: 243

Poster Number:

Non-Convex Primal-Dual Algorithm for Image Reconstruction from Physical Data in DECT (#2146)

B. Chen1, Z. Zhang1, D. Xia1, E. Y. Sidky1, X. Pan1, 2

1 The University of Chicago, Department of Radiology, Chicago, Illinois, United States of America
2 The University of Chicago, Department of Radiation and Cellular Oncology, Chicago, Illinois, United States of America


A first-order primal-dual Chambolle-Pock (CP) algorithm for convex optimization has been developed for CT and PET image reconstruction problems based on linear data models. Non-convex optimization problems, such as those based on a non-linear data model in multispectral CT, can not be solved by applying the CP algorithm. In this work, we propose a non-convex primal-dual algorithm, inspired by our previous work on the ASD-NC-POCS algorithm for the non-convex constrained TV-minimization problem in multispectral CT. The proposed non-convex CP (ncCP) algorithm can accommodate different optimization program designs including non-smooth objectives and/or constraints and also involves relatively fewer parameters. It is different than the MOCCA algorithm, which is also an extension of the primal-dual CP algorithm, in the sense that ncCP uses a global linear approximation to the data model and estimates the non-linear term locally, whereas MOCCA uses a local quadratic convex approximation to the data fidelity term. An algorithm instance of the proposed ncCP algorithm for a non-convex optimization problem in dual-energy CT has been derived and applied to physical head phantom data collected on a clinical scanner. Comparable basis and monochromatic images can be observed from the ncCP algorithm and the standard, data-domain decomposition method, in this preliminary study with two-full-rotation dual-energy data of overlapping rays. Future work will focus on the application of the algorithm to non-standard scanning configurations in multispectral CT, possibly without overlapping rays, and on the investigation of different optimization program designs, such as using KL divergence as the data fidelity function.

Keywords: dual-energy CT, non-convex optimization, primal-dual algorithm
Poster panel: 246

Poster Number:

Ultrasound Image Reconstruction Using Nesterov's Accelerated Gradient (#2203)

H. Wang1, B. Dalkilic1, H. Gemmeke2, T. Hopp2, J. Hesser1

1 Heidelberg University, Medical Faculty Mannheim, Experimental Radiation Oncology, Mannheim, Baden-Württemberg, Germany
2 Karlsruhe Institute of Technology, Institute for Data Processing and Electronics, Eggenstein-Leopoldshafen, Baden-Württemberg, Germany


The purpose of this paper is to investigate Nesterov’s accelerated gradient (NAG) method for the reconstruction of speed of sound and attenuation profiles in ultrasound computed tomography. Firstly, ultrasound (acoustic) wave propagation based on paraxial approximation was performed as the forward model. For iterative reconstruction, exact measurements were simulated from the forward model and then compared with the estimation measurements which were updated for each iteration based on the reconstructed profiles. This process is known as an inverse problem, which is tackled via minimizing the deviation between exact measurements and estimated measurements, i.e. via solving a nonlinear least-squares problem. To minimize this deviation, NAG was performed and compared with other optimization algorithms including gradient descent and Gauss-Newton conjugate gradient. Also, a line search method was used to change the step size for each iteration since finding proper step size was crucial for the convergence of such optimization algorithms. The strong Wolfe conditions were adopted as the termination condition for line search. Among the five tested algorithms (Gauss-Newton conjugate gradient, gradient descent, NAG, gradient descent with line search and NAG with line search), NAG with line search had fastest convergence rate (iteration number) compared to other methods. However, due to increased computational complexity of line search for each iteration, it requires extra computational time. On the other side, NAG with a fixed step size is the fastest method among all the tested method regarding computational time.

Keywords: Nesterov's accelerated gradient, ultrasound transmission imaging, USCT, inverse problem, line search
Poster panel: 249

Poster Number:

Contrast Recovery of Dual-Plane Breast PET with Feasible Region MLEM Reconstruction (#2311)

M. - W. Lee1, 2, M. - L. Jan1, 2, Y. - C. Ni3, J. - H. Hong2

1 Chang Gung University, Institute of Radiological Research, Taoyuan City, Taiwan
2 Chang Gung Memorial Hospital, Department of Radiation Oncology, Taoyuan City, Taiwan
3 Institute of Nuclear Energy Research, Health Physics Division, Taoyuan City, Taiwan


A dual-plane PET system can provide the benefits of flexible, low cost and high sensitivity, hence the dual-plane PET system is gradually applied to the breast cancer detections. However, the dual-plane PET with property of incomplete detection angles results in out-of-focus blurring in the reconstructed image. To overcome the image quality degradation caused by the limited detection angle problem, a reconstruction method of maximum-likelihood expectation-maximization with feasible region weighting (FRMLEM) is proposed. The out-of-focus blurring of dual-plane PET is affected by the limited detection angular coverage, hence the feasible region weightings related with the detection solid angles of line-of-responses and the voxel intensities were applied in the iterative procedure. In this study, both the simulations and measurements were performed to evaluate the contrast-recovery efficacy of FRMLEM. A simulated phantom was made up of a diameter 5-mm sphere which was inserted in a cylinder with radius 40 mm and length 160 mm. As for the measurement, a measured phantom consists of a 5 mm sphere immersed in a breast-shaped phantom with volume 600 mL. The contrast recovery percentage (CRP) was used to estimate the degree of contrast recovery that compared to the true contrasts in the simulations and measurements. From the reconstructed results of simulated phantoms with target/background (T/B) radioactivity ratios of 10 and 20 and with 0% and 15% Gaussian noises, the target intensities of FRMLEM were closer to the true intensities compared to those of MLEM. The reconstructed results of measured phantoms with T/B = 20 and 40 also show that the better CRPs can be obtained by FRMLEM. The CRPs of FRMLEM reconstructions in simulations and measurements were improved on average 2.2 times compared to those of MLEM. It is concluded that both the evaluations via simulations and measurements demonstrate the proposed FRMLEM have better capability of contrast recovery than MLEM.

Poster panel: 252

Poster Number:

A Sparse and Ergonomic Tomographic Image Reconstruction Technique based on Artificial Neural Networks (#2374)

M. - E. Tomazinaki1, 2, I. Lytrosyngounis1, E. Stiliaris1, 3

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


Considering the symmetry properties exhibited in the Radon Transform, a new type of tomographic image reconstruction technique based on conventional Feed-Forward type Artificial Neural Networks with fully equipped multilayer perceptrons is presented in this work. By introducing a simple remapping of the original sinogram in a constructive and sparse way without loosing the basic projective information, an ergonomically improved network architecture is here presented. The proposed prototype is free of any training sample dependencies or correlations. It provides comparable results to the standard reconstruction algorithms by reducing the computational time of the process. A variety of successfully reconstructed images with software phantoms but also with clinical SPECT data are shown and discussed.

Keywords: Artificial Inteligence, ANN, Radon Transform, Reduced Sinogram
Poster panel: 255

Poster Number:

Spatial-temporal constrained adaptive sinogram restoration for low-dose dynamic cerebral perfusion CT imaging (#2441)

Z. Bian1, 2, D. Zeng2, 3, T. Xie1, J. Huang2, 3, Q. Feng2, 4, J. Ma2, 3, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geveva, Switzerland
2 Southern Medical University, School of Biomedical Engineering, Guangzhou, China
3 Southern Medical University, Guangzhou Key Laboratory of Medical Radiation Imaging and Detection Technology, Guangzhou, China
4 Southern Medical University, Guangdong Province Key Laboratory of Medical Image Processing, Guangzhou, China


Dynamic cerebral perfusion computed tomography (PCT) imaging shows its potential in acute stroke diagnosis by visualizing and quantifying hemodynamic information of cerebral tissue and vessels. However, high radiation dose imposed on patients during dynamic PCT scanning has aroused growing concerns. To address this issue, we proposed a spatial-temporal constrained adaptive sinogram restoration (ST-ASR) method to improve the quality of dynamic cerebral perfusion CT imaging with the low-dose protocol. The proposed ST-ASR method explores the spatial-temporal correlation in dynamic cerebral PCT projections to develop a new spatial-temporal prior model for sinogram restoration and conducts an adaptive sinogram weighting according to the noise variance of sinogram data. Experimental results with clinical patient data demonstrate that the proposed ST-ASR method can achieve noticeable improvements over existing adaptive sinogram restoration method in terms of noise reduction and quantitative evaluations for low-dose dynamic cerebral PCT imaging.

Keywords: Dynamic cerebral perfusion computed tomography, adaptive sinogram restoration, low-dose
Poster panel: 258

Poster Number:

Spatially-Variant Image-Based Modeling of PSF Deformations with Application to Limited Angle Dual-Panel Breast-PET Data (#2520)

P. Gravel1, S. Surti1, S. Krishnamoorthy1, J. S. Karp1, S. Matej1

1 University of Pennsylvania, Department of Radiology, Philadelphia, Pennsylvania, United States of America


In this work we extend the functionality of the Direct Image Reconstruction for TOF (DIRECT) method to accommodate for spatially-variant image-based resolution modeling (IRM) of point-spread-function (PSF) deformations within reconstruction (RMvar). Additionally, we evaluate the spatially-variant IRM method in a post-reconstruction framework using Richardson-Lucy (RL) and Landweber (LW) deconvolution methods. The methods are applied on simulated dual-panel B-PET data for a breast phantom containing a set of lesions. Performance of the proposed methods is evaluated by comparing coefficient recovery coefficients (CRC) across lesions with those obtained with reconstruction of the breast phantom without IRM (noRM) and with a spatially-invariant IRM (fixRM). It is shown that our reconstruction method (i.e., RMvar) delivers considerably lower variation (i.e., %SD) of CRC across lesions with an improvement in mean of CRC when compared to both the noRM and fixRM reconstruction methods. In addition, when compared to the RMvar reconstruction, both post-reconstruction deconvolution approaches deliver results with slightly higher mean of CRC but at the cost of increased %SD of CRC across lesions, although the %SD is still considerably lower than for the noRM and fixRM reconstruction.

Keywords: Positron Emission Tomography, Resolution Modeling, Image Reconstruction, Image deconvolution, Dual-Panel Breast-PET
Poster panel: 261

Poster Number:

Reconstruction approaches for long axial field-of-view PET scanners (#2701)

M. E. Daube-Witherspoon1, V. Viswanath2, S. Matej1, S. Surti1, J. S. Karp1

1 University of Pennsylvania, Department of Radiology, Philadelphia, Pennsylvania, United States of America
2 University of Pennsylvania, Biomedical Engineering Department, Philadelphia, United States of America


The advantages of long axial field-of-view (AFOV) scanners include high sensitivity and multi-organ dynamic imaging. We are currently constructing a long AFOV whole-body PET scanner (PennPET Explorer) based on Philips Vereos detector tiles with multiple rings; to date, three rings have been assembled for a total of 70-cm AFOV. Long AFOV systems present challenges to quantitative image reconstruction such as (1) significantly increased computational, data storage, and memory demands; (2) non-negligible axial parallax blurring that increases with AFOV; and (3) non-uniform axial sensitivity across the AFOV. We have implemented two time-of-flight (TOF) reconstruction approaches that address these challenges in different ways: list-mode (LM) TOF ordered subsets expectation maximization (OSEM) used for clinical reconstructions and DIRECT, an efficient approach well-suited to long AFOV systems that takes advantage of the excellent TOF resolution (250 ps) of the PennPET Explorer to reduce the data sizes. To investigate the quantitative performance of the two approaches, we imaged the NEMA image quality (IQ) phantom on the PennPET Explorer. Additionally, we scanned the SNMMI Clinical Trials Network torso phantom (axial length: 30 cm) in one bed position on the multi-ring system. We also performed GATE simulations of the IQ and XCAT phantoms for a 70-cm AFOV system. The results of the simulations demonstrate the applicability of both reconstruction approaches to the long AFOV geometry. The measured results show that both approaches provide quantitative data from the PennPET Explorer. We will repeat these measurements and other clinically relevant phantom and patient studies on the 3-ring (70-cm AFOV) system to characterize the performance of each approach in terms of reconstruction speed, flexibility, and accuracy and precision over the range of activities that will be possible on this system.

Keywords: PET, total-body imaging
Poster panel: 264

Poster Number:

A Total Variance Regularization Method for Conductivity Imaging Using MR Phase (#2739)

X. Sun1, 2, L. Lu2, X. Liu1, W. Chen1, 2

1 University of Electronic Science and Technology of China, School of automation engineering, Chengdu, China
2 Southern Medical University, School of Biomedical Engineering, Guangzhou, China

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.


This paper proposed a total variance regularization method for conductivity imaging using Magnetic Resonance (MR) phase only. This method does not use the assumption that conductivity distribution is locally constant which is usually done. It improves the accuracy of conductivity imaging and is more robust against the noise.  Starting from the Maxell's equations, we can get a partial differential equation (PDE) including the gradients of conductivity. Through using finite difference method, the equation was transformed into a form of linear equations. Different from the previous method, it was solved by minimizing residues squares regularized by total variance. the proposed method does not need to give the boundary values of the region of interest (ROI) of an object and add an artificial diffusion term which blurs the conductivity map. The total variance term reduces the effect of noise and spurious oscillations in this method. The proposed method was compared with the conventional method which is using locally constant assumption. This method has been conducted in numerical simulations and human experiment. Simulation experiment results of the proposed method show lower reconstruction errors and more robust to the noise compared to conventional method. In human brain experiment, the result of proposed method shows more structures than the conventional method. The total variance regularization method for conductivity imaging improve the accuracy of conductivity maps and reduce noise, it has the potential for clinical applications.

Keywords: conductivity, MR EPT, phase based, magnetic resonance imaging
Poster panel: 267

Poster Number:

Improving Low Dose CT Performance Using TOF PET: Inherent Attenuation Information and Structure Similarity (#2841)

L. Cheng1, 2, Y. Liu1, 2, T. Ma1, 2

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


 In X-ray computed tomography (CT) examinations, patient dose reduction is essential for decreasing radiation risk. In clinical, CT dose can be reduced by lowering down tube current or acquiring less number of projections, at a cost of increased noise and artifacts. In modern TOF PET/CT systems, especially with the development of long axial field-of-view, high sensitivity PET, it is possible to improve CT image quality with the introduction of TOF-PET data. In this work, we propose a joint reconstruction method that uses both TOF-PET and low dose CT data. Joint total variation prior is introduced to further improve image quality. Different low-dose CT data acquisition methods and different PET count levels are considered in this study. Numerical simulation was conducted based on 2D NCAT phantom and a structural similarity (SSIM) index was employed for quantitative analysis. The results demonstrate that combined with joint prior, SSIM of joint reconstruction in flat regions (0.65~0.8) is significantly higher than SSIM of CT iterative reconstruction (<0.15). In regions including more details, SSIM of joint reconstruction (0.9~0.95) is higher than SSIM of CT iterative reconstruction with total variation regularization (0.75~0.9). When PET count level is very high, low dose CT can be improved even without prior term. In conclusion, joint reconstruction of low dose CT and PET data improves CT image quality. The fact that high PET emission count level leads to better low dose CT image quality demonstrates the potential of improving CT image quality in long axial-field-view PET systems.

Keywords: low dose CT, TOF PET, joint reconstruction
Poster panel: 270

Poster Number:

Image Quality Enhancement of Digital Breast Tomosynthesis Images by Deblurring with Deep Residual Convolutional Neural Networks (#1168)

Y. Choi1, H. Shim1, J. Baek1

1 Yonsei University, School of Integrated Technology, Incheon, Republic of Korea


Digital Breast Tomosynthesis (DBT) is widely used for early diagnosis of breast cancer due to its higher detectability compared to mammography. However, since the DBT acquires projection data within a limited angular range, blurring artifacts appears in reconstructed images. In this work, we proposed a method to reduce the blurring artifacts in DBT images by applying Deep Residual Convolutional Neural Networks (DRCNN), which were shown to be very effective for deblurring caused by camera motions. To evaluate the performance of the proposed method, we calculated the Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR) and structural similarity index (SSIM) between reference and output image from the proposed networks. The results show that the DRCNN reduces the blurring artifacts effectively, improving the  image quality.

Poster panel: 273

Poster Number:

Curvature Minimization for Edge-preserved Denoising in Tomographic Image (#1301)

Y. Kim1, S. Cho1, K. Ly1

1 KAIST, Nuclear and Quantum Engineering, Daegeon, Republic of Korea


In medical imaging, noise reduction is a challenging issue because fine details with important diagnostic information can also be adversely affected during the denoising procedure. In this paper, we introduced a curvature minimization method to medical image processing for noise removal. The properties of curvature minimization are contrasted with conventional total variation minimization scheme. In addition, we implemented several mathematically well-defined curvature minimization methods to discuss their pros and cons.

Keywords: denoising, curvature minimization, medical image processing, gauss curvature, mean curvature
Poster panel: 276

Poster Number:

Exploring the association between multiparametric MRI and hypoxia in prostate cancer (#1496)

Y. Sun1, H. M. Reynolds2, D. Byrne2, S. Williams2, D. Wraith3, C. Mitchell2, D. Murphy2, A. Haworth1

1 The University of Sydney, Institute of Medical Physics, Sydney, Australia
2 Peter MacCallum Cancer Centre, Melbourne, Australia
3 Queensland University of Technology, Brisbane, Australia


Oxygen is a radio-sensitiser in radiotherapy (RT) and hypoxia (the lack of oxygen) may influence the efficacy of RT. Hence accurate estimations of hypoxia, preferably conducted non-invasively, are essential to optimise personalised RT. This exploratory study aimed to investigate the association between multiparametric MRI (mpMRI) data and hypoxia in prostate cancer using genetic profiles as a surrogate. In vivo mpMRI was acquired from six patients prior to radical prostatectomy. Sequences included T2-weighted (T2w) imaging, diffusion-weighted imaging (DWI), dynamic contrast enhanced MRI (DCE-MRI) and blood oxygen-level dependent (BOLD) imaging. Texture features (n=51) and parametric maps (n=5) were computed from T2w and functional MRI data respectively. After surgery, punch biopsies were performed on prostate tissue blocks. The hypoxia levels of the tissue blocks were assessed by obtaining full transcriptome genetic profiles using next generation sequencing. Pearson correlation coefficients were calculated between mpMRI data and hypoxia-related gene expression levels. This identified 34 candidate imaging features, 6 were from the mpMRI data and the remaining 28 were T2w texture features. Further validation using glucose transporter 1 (GLUT-1) immunohistochemistry (IHC) showed that 16 of 28 T2w texture features achieved weak but significant correlations (p < 0.05). This indicated the feasibility of combining these weak predictors into a classifier to non-invasively assess hypoxia status in prostate cancer. In summary, this is the first study using radiogenomics approach to investigate the correlation between mpMRI data and prostate hypoxia.

Keywords: MRI, hypoxia, prostate cancer, gene sequencing, radiogenomics
Poster panel: 279

Poster Number:

Use of a 4D Deep Autoencoder to Denoise Dynamic PET Data (#1607)

I. S. Klyuzhin1, J. - C. Cheng2, C. W. J. Bevington2, V. Sossi2

1 University of British Columbia, Department of Medicine, Division of Neurology, Vancouver, British Columbia, Canada
2 University of British Columbia, Department of Physics and Astronomy, Vancouver, British Columbia, Canada


Positron emission tomography (PET) studies rely on kinetic modeling (KM) to estimate intrinsic tissue parameters, such as the non-displaceable binding potential. Prior to use in KM, reconstructed voxel time-activity curves (TACs) must be often denoised. Application of smoothing filters generally reduces noise but also degrades contrast and may introduce bias; better TAC denoising methods can increase image quality, accuracy, and reproducibility.

In this work, we explore the use of deep denoising autoencoders (DAE) for denoising of PET TACs. DAEs represent a class of deep neural networks that are trained to recover the true signal from its noisy version. DAEs have been shown to achieve state-of-the-art signal denoising in many areas; thus, we hypothesize that DAEs may be able to achieve better denoising of PET TACs compared to other methods. 

We construct a multi-layer DAE that operates on 4D data consisting of stacked TACs from neighboring voxels. We simulate ground truth (GT) and reconstructed (noisy) dynamic images, and use them for DAE training and testing. Dynamic images denoised by a trained DAE are compared to the GT, and to the images denoised by other commonly used methods (temporal averaging, Gaussian smoothing, 4D smoothing/deconvolution). 

The DAE-processed test image had higher contrast and lower noise compared to other methods. The measured mean absolute error (MAE) of TACs, relative to the GT, was 2.0x103 Bq/cc with smoothing/deconvolution (best reference method), and 1.7x103 Bq/cc with DAE. Background noise in low-count frames, measured as the coefficient of variation was 28% with smoothing/deconvolution, and 22% with DAE. Thus, the DAE produced the highest quality post-processed dynamic images. This is likely achieved as a result of the deep network learning the probable signal distribution as well as the noise properties of the TACs. Further improvement is expected to be achieved with the DAE hyperparameter and architecture optimization.

Keywords: PET, denoising, autoencoder, deep learning, kinetic modeling
Poster panel: 282

Poster Number:

Multi-planar reconstruction under incomplete data by enhanced fuzzy radial basis function neural networks (#1665)

Z. Chao1, D. Kim2, H. - J. Kim1, 2

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


Slice images based on single section cannot be commonly satisfied with the clinician's diagnosis or observation. Therefore, it’s necessary to simulate required slices by multi-planar reconstruction technology. In the case of complete data, it’s not hard to get the series of clear images from other sections. In addition, for the case of incomplete data, the interpolation methods are commonly adopted to make up for lost data based on reconstructed images. However, the effects were not ideal. In this study, we propose a new method of improving interpolation algorithm by enhanced fuzzy radial basis function neural network. At first, a series of incomplete transverse section images which have been accurately registered are adopted. Then we superpose sequence images to get three-dimensional data volume. So, we can get coronal or sagittal images by reformatting data volume. Based on this, for a certain reconstructed image, bicubic interpolation was applied to compensate for the lost data. Subsequently, we repair the interpolated data by the combination of fuzzy radial basis function neural network. The system included five layers: input, fuzzy partition, front combination, inference, and output. Here, we selected 2 pixels which are neighbor to a certain lost data as inputs of our system and used interpolated data (the same position as that lost data) which as training sample, then we trained the system by Gravitational search algorithm. Based on this, we used 15 sets of input data (all data were selected around the interpolated data) to get 15 output data, then the final output data were obtained by inverse distance weighted algorithm. At last, we repaired the whole interpolated data. In this experiment, we used one group of datasets: Brain MRI image. Finally, subjective observation and objective evaluation prove the superiority and effectiveness of proposed method comparing with interpolation method.

Keywords: multi-planar reconstruction, enhanced fuzzy radial basis function neural network, Gravitational search algorithm
Poster panel: 285

Poster Number:

Improvement of image resolution via a deep convolutional generative adversarial network (DCGAN) in a photon-counting CT system (#1780)

S. Choi1, D. Lee2, D. Kim2, H. Kim1, H. - J. Kim1, 2

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


A photon-counting detector challenges for use of a large number of pixels due to an expensive unit price per each pixel. Thus, it is difficult to expect a diagnostic image quality at current lab-based photon counting technology no matter how good the detector resolution is. This work describes an efficient way to improve a low-resolution into a high-resolution image using a deep convolutional generative adversarial network (DCGAN). A total number of 4 images of two different cylindrical targets acquired from low- and high-energy bins which have a size of 64×64 reconstructed pixels array were input to generate 256×256 size images using the proposed deep network. A fake output of high-resolution image was compared with the same size of images produced by conventional bilinear and cubic interpolation processed images. A total number of 192 axial CT scanned images of Catphan® phantom were used for the training set. The results indicated that the images performed with the proposed deep learning-based scheme restored the low-resolution images well by providing the peak signal-to-noise ratios (PSNRs) of 39.5-63.2 with respect to the original images in our validation set, which was the highest values among the conventional bilinear and cubic interpolation schemes. Consequently, the proposed DCGAN image processing method could provide an efficient way to improve the image resolution with high perceptual quality.

Poster panel: 288

Poster Number:

Back Propagation Neural Network and 18F-Florbetapir PET for Early Detection of Alzheimer's Disease (#1853)

I. Ozsahin1, G. Altinoglu1, B. Sekeroglu2

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


Alzheimer’s disease (AD) is the most common type of dementia, characterized by progressive loss of neural cells, with no clinical treatments to avoid the manifestation or progression of the disease. This aspect of treatment failure is considered as the result of late diagnosis and thus late intervention, highlighting the essential need to develop early and accurate detection methods. Importantly, the use of positron emission tomography (PET) imaging, with probe 18F-florbetapir, to image amyloid beta (Aβ) plaques formed from the aggregation of Aβ protein is a good candidate as an early detection technique since Aβ aggregation is considered as the “start” of the degenerative process that manifests years before the clinical symptoms appear. Thus, it can be used for assessing the amyloid load in vivo across a continuum from a healthy aging brain to AD. Yet, classification of clinical data for AD is challenging due to its highly heterogeneous nature. In the current study, we propose the use of back propagation neural networks together with 18F-florbetapir PET data taken from Alzheimer's Disease Neuroimaging Initiative (ADNI) database for an automated classification of four groups including AD, mild cognitive impairment (MCI-prodromal stage of AD; as late-LMCI and early-EMCI), and significant memory concern (SMC) versus normal control (NC). The results show that the automated classification of AD produced high accuracy measure of 87.9%. Also, the results for the prodromal stages of the disease are promising: 66.4%, 60.0%, and 52.9% for LCMI, EMCI, and SMC, respectively, and indicating that our method has potential for very early detection of AD.

Keywords: 18F-florbetapir, ADNI, Alzheimer's disease, neural network, PET
Poster panel: 291

Poster Number:

Spatially-guided non-local mean filter for denoising of clinical whole-body PET images (#1873)

H. Arabi1, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Genève, Switzerland


A novel spatially-guided non-local mean (SG-NLM) filter for denoising of clinical whole-body PET images is proposed. In this approach, PET images are initially clustered into a number of regions with homogeneous uptake based on which a Bayesian maximum a posteriori framework is exploited to estimate the noise level for automatic setting of smoothing parameters The conventional NLM filter defines a search window to limit the exhaustive search required to find similar patches. In the SG-NLM approach, instead of defining a search window, only patches belonging to the same regions are compared to facilitate the patch searching process without additional computational burden. The SG-NLM filter was evaluated through comparison with conventional post-reconstruction Gaussian and NLM filters. Experimental measurements using the Jaszczak phantom and clinical whole-body PET/CT studies were used to assess the performance of the proposed PET denoising technique. For the experimental Jaszczak phantom, the signal to noise ratio (SNR) improved from 26.5 when using Gaussian smoothing to 29.9 and 30.7 using the NLM and SG-NLM filters, respectively. The clinical studies further demonstrated the superior performance of the SG-NLM filter, yielding a quantification change in malignant lesions of -2.4%, with respect to unfiltered images, compared to -11.7% and -2.9% achieved using Gaussian and NLM filters, respectively. The SG-NLM approach proved to improve the contrast, SNR and quantitative accuracy compared to Gaussian and NLM filtering approaches, and can be utilized as an alternative post-reconstruction filter in clinical whole-body PET/CT imaging.

Keywords: whole-body PET
Poster panel: 294

Poster Number:

3D-CNN HadNet classification of MRI for Alzheimer’s Disease diagnosis (#1912)

I. Sahumbaiev1, J. Ramírez1, A. Popov1, J. M. Górriz1, A. Ortiz1

1 NTUU , Biomedical Engineering, Kyiv, Ukraine


Alzheimer’s disease (AD) still remains without an effective treatment, and its clinical diagnosis largely depends on the experience of clinicians. In this paper, we developed a learning system that employs a deep 3D convolutional neural network (3D-CNN) in order to learn and discriminate AD features in Magnetic Resonance Imaging (MRI) data. MR images were collected from Alzheimer’s Disease Neuroimaging Initiative (ADNI) and spatially normalized with statistical parametric mapping (SPM) toolbox. The proposed 3D-CNN architecture (HadNet) features inception approach, and its hyper-parameters were tuned through the Bayesian optimization process. The experiments assessed the accuracy of the HadNet architecture in a three-class classification problem to discriminate between healthy subjects, mild cognitive impairment (MCI) and AD patients. After optimization of the network architecture and hyper-parameters, we achieved the final specificity of 83% and sensitivity of 86% with classification accuracy of 84%.

Keywords: Alzheimer’s disease, MR images, Convolutional Neural Network, 3D-CNN, Deep learning
Poster panel: 297

Poster Number:

Improved separation of Alzheimer's disease and related disorders using dual-point amyloid-PET (#1908)

F. Segovia-Roman1, J. M. Górriz1, J. Ramírez1, D. Castillo-Barnes1, F. J. Martínez-Murcia1, D. Salas-Gonzalez1, R. Sánchez-Vañó2, P. Sopena-Novales2, M. Gomez-Rio3

1 University of Granada, Department of Signal Theory, Networking and Communications, Granada, Spain
2 9 de Octubre Hospital, Department of Nuclear Medicine, Valencia, Spain
3 Virgen de las Nieves University Hospital, Department of Nuclear Medicine, Granada, Spain


18F-Florbetaben (FBB) is an amyloid radiotracer that is being increasingly used to assist the diagnosis of Alzheimer’s disease (AD). Recent studies have suggested that early acquisitions (immediately after the radiotracer injection) of 18F-FBB data provide information about the neuronal injury. Although this assumption still needs to be corroborated by further studies with larger datasets, this would allow obtaining information about the brain injury and amyloid deposits in single examination, using a dual-point protocol. Despite previous works are focused on analyzing the equivalence between early 18F-FBB-PET and 18F-FDG-PET images, in this work we analyzed whether the information provided by early and standard 18F-FBB-PET acquisitions is complementary and can be used together to improve the automatic separation of AD and non-AD patients. Two approaches were proposed to combine the two images from each patient into a single observation: feature concatenation and multiple kernel learning. Then, a Support Vector Machine classifier was used separate the groups. The classification performance was estimated using a dataset with 18F-FBB-PET data from 80 patients (44 AD and 36 non-AD) along with a cross-validation scheme. Accuracy, sensitivity and specificity achieved by the proposed approaches were compared to those obtained by approaches using only data from one acquisition. The results suggest that using both 18F-FBB-PET acquisitions improves the automatic separation of AD and non-AD patients. They also corroborate that early 18F-FBB-PET images contains useful information (different to that contained in standard 18F-FBB-PET images) to distinguish between AD and non-AD patients. That would make early 18F-FBB-PET images an affordable way to obtain additional information from amyloid tracers that would result in more accurate diagnosis.

Keywords: 18F-Florbetaben-PET, Alzheimer's disease, Support Vector Machine, Multiple Kernel Learning, Multimodal analysis
Poster panel: 300

Poster Number:

Improved coordinate reconstruction for SiPM based gamma detector with pixelated scintillating crystal and multiplexed readout (#2065)

H. Poladyan1, O. Bubon2, A. Teymurazyan3, S. Senchurov4, A. Reznik5

1 Lakehead University, Biotechnology PhD Program, Thunder Bay, Ontario, Canada
2 Lakehead University, Chemistry and Materials Sciences , Thunder Bay, Ontario, Canada
3 University of Regina , Faculty of Science, Regina, Saskatchewan, Canada
4 Taras Shevchenko National University of Kyiv, Molecular Physics, Kyiv, Ukraine
5 Lakehead University, Physics Department, Thunder Bay, Ontario, Canada


Currently the prevailing high-resolution Positron Emission Tomography (PET) detector architecture consists of a high-density and high-atomic number scintillating crystals coupled to a pixelated photosensitive layer such as position sensitive photomultiplier tubes (PS-PMTs) or silicon photomultiplier (SiPM) arrays. In more common designs the scintillating crystal itself is segmented into two dimensional array, in which case light spreading layer is introduced between the finely segmented scintillating crystal array and coarsely segmented photosensitive array in order to disperse the light from scintillation over several elements of the photosensitive layer. Such detectors typically rely on some form of analog signal multiplexing in order to minimize the number of readout electronics channels and the positions of the gamma ray interactions are calculated using simple Centre-of-Gravity (CoG) approach. This logic has non-linear response for the events incident at the periphery of the detector and introduces systematic error in calculations of gamma ray interaction coordinates resulting in field-of-view (FOV) compression and resolution degradation. Here we report on applicability of the Truncated Center of Gravity (TCoG) and Raised To the Power (RTP) methods for reconstruction of the positions of the gamma ray interactions as applied to a newly developed small-scale PET detector prototype based on 24x24 LYSO scintillation crystals and 8x8 SiPM array with multiplexed readout. A notable advantage of using TCoG algorithm instead of CoG is demonstrated, with significant improvement in reconstruction of coordinate for gamma ray interaction point. Usage of TCoG increases spatial resolution, while decreasing the effect of FOV shrinkage due to CoG algorithm, making it possible to resolve all pixels in crystal array and perform coordinate correction and pixel-by-pixel energy correction and/or filtration.

Keywords: PET, SiPM, CoG, TCoG, RTP
Poster panel: 303

Poster Number:

Improving Lung Lesion Detection in Low Dose Positron Emission Tomography Images Using Machine Learning (#2156)

Y. Nai1, J. Schaefferkoetter1, D. A. Fakhry-Darian1, M. Conti2, D. W. Townsend1, 5, I. Tham1, 4, D. C. Alexander1, 3, A. Reilhac1

1 Clinical Imaging Research Center, Imaging, Singapore, Singapore
2 Siemens Healthcare, Molecular Imaging, Knoxville, Tennessee, United States of America
3 University College of London, Centre for Medical Image Computing and Department of Computer Science, London, United Kingdom
4 National University Cancer Institute , Department of Radiation Oncology, Singapore, Singapore
5 National University Hospital, Department of Diagnostic Radiology, Singapore, Singapore


Lung cancer suffers from poor prognosis, leading to high death rates. Combined PET/CT improves lung lesion detection but requires low dose protocols for frequent disease screening and monitoring. In this study, we investigate the feasibility of using machine learning to improve low dose PET images to standard dose, high-quality images for better lesion detection at low dose PET scans. We employ image quality transfer (IQT), which is a machine learning algorithm that employs patch-regression to map parameters from low to high-quality images e.g. enhancing resolution or information content. We acquired 20 standard dose PET images and simulated low dose PET images with 9 different count levels from the standard dose PET images. For each count levels, 10 pairs of standard dose PET images with one simulated low dose PET images were used to train linear, single non-linear regression tree, and random regression-forest models for IQT. The models were then used to estimate standard dose images from low dose images for each count levels for 5 different subjects. Improvement in image quality and lesion detection could be observed in the images estimated from the low dose images using IQT. Among the models employed, regression tree model produced the best estimates of standard dose PET images. An average bias of less than 20% in SUVmean of 90 lesions in the estimated images from the standard dose PET images can be obtained down to 7.5 x 106 counts. Improvement in image quality shows the potential of IQT in improving the accuracy in lung lesion detection at low dose PET imaging. We planned to further optimize the IQT algorithm to reduce SUV bias.

Poster panel: 306

Poster Number:

The preliminary study of the de-noising method in the phase contrast imaging based on the convolutional neural network (#2186)

S. Lee1, O. Oh1, Y. Kim1, S. W. Lee1, G. Wang2

1 Pusan National University, school of mechanical engineering, Busan, Republic of Korea
2 Rensselaer Polytechnic Institute, Biomedical Engineering, Troy, New York, United States of America


An x-ray Talbot-Lau grating interferometer is widely used for phase contrast imaging. It has three gratings between x-ray and detector, and the phase contrast images are measured by the phase stepping method which translates one of the gratings. Increasing the number of phase stepping is advantageous for more accurate phase-shift information in the images. However, the increased phase stepping number means higher exposure time in the system, and it is inefficient for medical and industrial applications. In this study, we suggest an up-sampling method by the one-dimensional convolutional neural network to reduce the exposure time and improve signal to noise ratio (SNR). The general interferogram from the Talbot-Lau grating interferometer is generated usually by five to ten phase stepping. We have built a neural network architecture consisting of several layers of one-dimensional convolution and transposed convolution. For the training, we setup an x-ray Talbot-Lau grating interferometer and generated training data sets. We used the phase stepping data of five sampling points as a training data, and the phase stepping data of twenty sampling points as a ground truth.  Thus, the up-sampling factor was four, and the whole process for the learning was proceeded based on the Tensorflow. For the learning, four convolution layers and four transposed convolution layers were built. Each layer was activated via soft-sign function. The optimal learning rate in this study was 1e-3. From the results, we have found that the convolutional neural network well estimates the phase shift in the Talbot-Lau grating interferometry.

Keywords: X-ray, Talbot-Lau grating interferometer, convolutional neural network, up-sampling, de-noising
Poster panel: 309

Poster Number:

Radiomics-based prediction of symptomatic intracerebral hemorrhage before thrombolysis therapy in unenhanced CT imaging (#2390)

H. Xu1, 2, W. Lyu1, 2, D. Zeng1, 2, Z. Bian1, 2, J. Huang1, 2, Z. Deng3, 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
3 Southern Medical University, Department of Neurology, NanFang Hospital, Guangzhou, China

This work was supported in part by the National Natural Science Foundation of China under Grant No. 61571214, the Guangdong Natural Science Foundation under Grant No. 2015A030313271, the Science and Technology Program of Guangdong, China under Grant No. 2015B020233008, the Science and Technology Program of Guangzhou, China under Grant Nos. 201510010039 and 201705030009.


Symptomatic intracerebral hemorrhage (sICH) is rare but the most devastating complication of thrombolysis therapy for acute ischemic stroke. Early prediction of sICH is thus critical for an effective and safe thrombolysis therapy. This study aims to predict the probability of sICH before thrombolysis therapy via radiomics analysis on unenhanced CT images. A total of 300 patients from three different centers were respectively enrolled (118 patients from NanFang Hospital-NF, 99 patients from Huadu District People’s Hospital-HD, and 83 patients from Fuzhou General Hospital-FG). All the unenhanced CT images were acquired within 8 hours of symptom onset. There are 23 (7.6%) patients developed as sICH after therapy (6, 5, 12 patients for the three centers, respectively). 217 patients from the two centers of NF and HD were used for training, and 83 patients from FG center were used as independent validation cohort. In this study, a total of 1114 and 434 slices CT images were selected to construct training and validation cohorts, respectively. 538 radiomics features were extracted from each slice, and slice-based minimum redundancy and maximum relevance (mRMR) feature selection in combination with random forest (RF) classifier were adopted to construct the prediction model. The top five significant radiomics features (GLCM-LL-Correlation, GLCM-HH-Homogeneity, Histogram-Kurtosis, shape-PAratio, and NGTDM-HH-Contrast) are incorporated to construct the final model, resulted in the area under the receiver-operating characteristic curve (AUC-ROC) of 0.74 and 0.71 in training and validation cohorts, respectively. The radiomics analysis on multi-center pre-therapy unenhanced CT images demonstrated potential to the early prediction of sICH in acute ischemic stroke.


Keywords: Symptomatic intracerebral hemorrhage, Radiomics, thrombolysis therapy
Poster panel: 312

Poster Number:

A New Look of Gray-scale Co-occurrence for Multi-scale Texture Measures with Application to Classify Colorectal Polyps via CT Colonography (#2524)

W. Cao1, Z. J. Liang1, M. Pomeroy1, P. J. Pickhardt2

1 Stony Brook University, Department of Radiology, Stony Brook, New York, United States of America
2 University of Wisconsin Medical School, Department of Radiology, Madison, Wisconsin, United States of America


Classifying colorectal polyps is clinically important but technically challenging. Haralick texture features from gray-scale co-occurrence (GSC) texture measures have shown the potential to relive the challenging. This study aims to increase the potential by exploring GSC multi-scale texture measures and minimizing texture measure spatial variation. The first scale is along two directions of x-axis (0 degree) and y-axis (90 degree). The second scale is along another two directions of 45 and 135 degrees. The third scale is along four directions of 30, 60, 120 and 150 degrees. Higher order scales are determined in the same manner by angular sampling around a concerned pixel and grouping those directions into a new scale according to their same distance from the concerned pixel to the first hit pixel along the directions during the angular sampling. Texture measures from each scale are analyzed and all scales are ranked according their discriminating power by machine learning. The ranked multi-scale textures are applied to classify the polyps in a database including 32 invasive adenocarcinoma and 31 benign adenomas. The Haralick texture measures predicted the benign lesions by AUC score of mean 0.9535 and standard deviation 0.0100 as baseline for comparison. The presented multi-scale texture measures improved the prediction power to 0.9824 with standard deviation 0.0085. The innovation of integrating multi-scale texture measures and optimal angular sampling rate through the lesion space not only significantly improved the discriminating power in terms of AUC score, but also noticeably minimized the spatial variation in terms of standard deviation.

Keywords: Computed tomographic colonography, polyp characterization, multi-scale gray-scale co-occurrence, texture measures, colon cancer
Poster panel: 315

Poster Number:

Material Decomposition of Spectral CT by AUTOMAP   (#2574)

Z. Chen1, L. Li1

1 Tsinghua University, Beijing, China


AUTOMAP is an end-to-end image reconstruction method using neural network. This method has achieved good results in four common imaging methods. Inspired by this, we design a method to generate phantoms from natural image datasets, and apply the AUTOMAP network to match the relationship between the duel-energy reconstruction image and the volume fraction distribution of scanned object. However, the simulation result is not satisfying. The outputs are blurred, and their MSE (mean square error) is larger than traditional methods. Some improvements are proposed to further study.

Keywords: material decomposition, spectral CT, artificial neural network, deep learning
Poster panel: 318

Poster Number:

A Photon-Counting Detector Calibration Method  (#2675)

X. Li1, L. R. Furenlid1

1 University of Arizona, Radiology Research Laboratory, Tucson, Arizona, United States of America


We have developed a fast detector calibration method, which is capable of calibrating photon-counting detectors quickly and conveniently, and potentially has DOI calibration capability. Compared with traditional calibration methods, this calibration technique doesn’t require a finely collimated pencil beam, and can calibrate a photon-counting detector with reduced time. The unique part of this technology is its data-processing method called common-data subset (CDS) method, which is used to synthesize pencil gamma-ray beams or even single points for detector calibration. If using fan beams, only 2N scans (in x and y directions) are necessary to calibrate the whole detector surface instead of doing N^2 scans (if using pencil gamma-ray beam). For calibration of the DOI information, 3N scans are necessary, within which N scans are performed with slant beams. Moreover, this calibration technique can also work for detectors with complicated mean-detector-response functions (MDRFs). Both Monte-Carlo simulations and experiment results support the feasibility of this method, which enables simple, fast and robust calibration for a variety of photon-counting detectors. 

Keywords: DOI, calibration, detector, photon-counting, digital signal processing
Poster panel: 321

Poster Number:

Optimisation of a Novel Pipeline for 3D Reconstruction of 2D Serial Brain Sections and Proof-of-concept with Autoradiography Data (#2877)

G. Emvalomenos1, 2, M. Safavi-Naeini1, K. Bambery1, L. Zimmer3, 4, M. - C. Grégoire1, 2, A. Reilhac5, F. Chauveau4, A. Charil1, 2

1 Australian Nuclear Science and Technology Organisation, Lucas Heights, Australia
2 The University of Sydney, Brain & Mind Centre, Sydney, Australia
3 Univ. Lyon1, Lyon Neuroscience Research Center; CNRS UMR5292; INSERM U1028, Lyon, France
4 CERMEP, Imagerie du vivant, Lyon, France
5 NUS A*STAR Clinical Imaging Research Centre, Singapore, Singapore


Autoradiography provides 2D images of the binding of a radiotracer, with a higher resolution than in vivo Positron Emission Tomography. However, 3D tissue conformation is lost during tissue sectioning. We present a novel pipeline for 2D-to-3D reconstruction, optimized for autoradiography only sections and designed for routine use. It includes a preprocessing step with GNU Octave or Matlab (Mathworks) and a registration algorithm developed in Perl and based on ‘minctools’ (www.bic.mni.mcgill.ca). We optimized our pipeline on test datasets created from a 3D MRI mouse brain template. The volume was sliced along the coronal plane into 60µm sections. We applied realistic random rotations, translations and 2D-scaling to each section and added typical artefacts occurring during cryo-sectioning such as intensity changes, bubbles and tears., The iterative processing of each section consists of: 1) center of mass registration, 2) rigid registration (intensity-based information with cross-correlation similarity measure), and 3) local non-linear transformation. 2D registration accuracy was quantitatively assessed using the normalized correlation coefficient (NCC) for every two adjacent section-pair. We used the mean square error (MSE) between NCC values from the initial template and from those of the reconstructed volumes. The MSE values decreased by a factor of 500 following the full iterative pipeline; demonstrating a very high agreement of the reconstructed volume with the template. This was supported by a visual evaluation of the recovered internal structures by a specialist. In addition to a proof-of-concept of a single modality 3D reconstruction pipeline, we present a methodology to validate this pipeline for multiple radiotracers and an application to sections obtained with a radiotracer targeting myelin.

Keywords: autoradiography, 3D reconstruction, 2D registration, serial sections, MRI template
Poster panel: 324

Poster Number:

A Beam-Hardening Correction Method Based on a Discrete Spectral Attenuation Model in X-ray Microcomputed Tomography (#1082)

D. Lee1, J. Soh1, S. Lim1, S. Cho1

1 KAIST, Nuclear and Quantum Engineering, Daejeon, Republic of Korea


Tube-based X-ray has polychromatic energy spectrum. In microcomputed tomography (μCT), the beam-hardening effect is relatively severe for it uses rather low-energy X-rays. It can cause quantitative errors and image artifacts in the reconstructed tomographic images. The aim of this research is to develop an efficient method for beam-hardening correction, by modeling a discrete spectral attenuation. The model is based on an assumption that the measured energy intensity can be approximated by a summation of discrete energy intensities. The feasibility of this method was demonstrated on datasets of several samples obtained by GATE simulation. The results show that the beam-hardening artifacts can be substantially removed.

Keywords: X-ray, microcomputed tomography, beam-hardening effect
Poster panel: 327

Poster Number:

Generation of attenuation map for whole-body PET/MRI using a deep neural network applied to simultaneously reconstructed activity and attenuation and Dixon MRI (#1210)

D. Hwang1, 2, K. Y. Kim1, 4, S. K. Kang1, 2, S. Seo3, J. S. Lee2, 4

1 Seoul National University, Department of Biomedical Sciences, Seoul, Republic of Korea
2 Seoul National University, Department of Nuclear Medicine, Seoul, Republic of Korea
3 Gachon University, Department of Neuroscience, Incheon, Republic of Korea
4 Brightonix Imaging Inc., Seoul, Republic of Korea


Attenuation maps used in current whole-body PET/MRI studies have limited accuracy compared to PET/CT systems. Various techniques have been proposed to generate accurate attenuation maps in PET/MRI including Dixon sequence based four-segment approach (water, fat, lung, and air) and maximum likelihood reconstruction of activity and attenuation (MLAA) algorithm, one of simultaneous reconstruction algorithms. However, these methods suffer from lack of bone information in Dixon sequence and high noise level in MLAA attenuation, causing considerable quantification errors.

In this study, we propose a new deep learning based approach to PET/MRI attenuation correction. In this approach, we use MLAA activity and attenuation maps and four-segment map that is corresponding to Dixon MRI-based attenuation map as inputs to a convolutional neural network (CNN) to learn CT-derived attenuation map for more accurate whole-body PET/MRI attenuation correction than currently available Dixon-only four-segment method.

The CNN generates less noisy attenuation maps than MLAA and achieves better bone identification than Dixon MRI. The average Dice similarity coefficient for bone region between μ-CNN and μ-CT was 0.68 that was significantly higher than that between μ-MLAA and μ-CT (0.37). Also, the CNN results show the best pixel-by-pixel correlation with the CT-based ones and remarkably reduced absolute and percent difference of activity maps from the CT-based attenuation correction.

In conclusion, we have developed a deep neural network that successfully integrated the information obtained from the simultaneous activity and attenuation reconstruction and 4-segment map based on Dixon MRI to produce a more reliable attenuation map for 511 keV photon than conventional Dixon MRI method.

Keywords: PET, PET/MRI, Simultaneous reconstruction, Deep learning, Attenuation correction
Poster panel: 330

Poster Number:

A method to reduce data-redundancy artifacts for arbitrary source trajectories in CT imaging  (#1300)

T. Sun1, 4, R. Fulton2, 3, J. Nuyts1

1 KU Leuven, Nuclear Medicine & Molecular Imaging, Leuven, Belgium
2 University of Sydney, Faculty of Health Sciences, School of Physics, Sydney, Australia
3 Westmead Hospital, Department of Medical Physics, Westmead, Australia
4 Harvard Medical School, Massachusetts General Hospital, Boston, United States of America


While circular and helical scanning trajectories are mostly used in CT imaging, sometimes irregular source trajectories are also present. These include the intended ones which aim at improving the reconstructed image quality, and the unintended ones which are typically resulted from a motion correction process. The usage of these irregular trajectories can sometimes cause a problem called data-redundancy, i.e. different voxels of the object are measured for different times. Failing to handle the data-redundancy may create artifacts in the reconstructed images. We proposed an approach which can correct these artifacts effectively within the acceptable amount of time. The approach are based on two assumptions: (1) a (converged) iterative reconstruction can produce an image free from artifacts due to the redundancy; (2) these artifacts are mostly at low-frequency. Based on these assumptions, we were able to extract the artifacts-only image, and that image can be subtracted from the uncorrected image subsequently. The approach was validated for a motion-corrected patient scan, which is contaminated by the patient movement. Results showed that most of the artifacts were successfully removed in the image after correction. Further evaluations are required for the scans acquired with other irregular trajectories.

Keywords: CT imaging, artifacts redution
Poster panel: 333

Poster Number:

Novel deep learning-based CT synthesis algorithm for MRI-guided PET attenuation correction in brain PET/MR imaging (#1567)

H. Arabi1, G. Zeng2, G. Zheng2, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Switzerland
2 University of Bern, Institute for Surgical Technology and Biomechanics, Bern, Switzerland


MRI-guided synthetic CT (sCT) generation is one of the main challenges hampering quantitative PET/MR imaging and MRI-only radiation planning. Deep learning-based approaches have recently gained momentum in a variety of medical imaging applications. In this work, a novel synthetic CT generation algorithm based on deep convolutional neural network is proposed for MRI-guided attenuation correction in PET/MRI. The proposed algorithm (AsCT) exploits adversarial semantic structure learning implemented as a CT segmentation approach to constrain the synthetic CT generation process. The proposed technique was compared to an atlas-based method (Bone-Atl), previously developed for MRI-only radiation planning, as well as the commercial segmentation-based approach implemented on the Philips TF PET/MRI system. The evaluation was performed using clinical brain studies of 11 patients who have undergone PET/CT and MRI scanning. The accuracy of CT value estimation and cortical bone identification were assessed for the three different methods taking CT images as reference. Bias of tracer uptake (SUV) was measured on attenuation corrected PET images using the three techniques taking CT-based attenuation corrected PET as reference. Bone-Atl exhibited slightly better cortical bone extraction resulting in Dice coefficient (DSC) of 0.88±0.02 compared to AsCT (DSC=0.85±0.04). AsCT and Bone-Atl methods performed similarly in terms of accuracy of CT value estimation where a mean absolute error of about 116 HU was obtained for the whole head region. Quantitative analysis of brain PET images demonstrated competitive performance of Bone-Atl and AsCT methods where mean relative errors of -1.8% and 4.1% were achieved in bony structures, respectively, while the segmentation-based approach led to a mean SUV error of -18.8%. The proposed AsCT algorithm showed competitive performance with respect to the atlas-based method and outperformed the segmentation-based method with clinically tolerable errors.

Keywords: PET/MRI, deep convolutional neural network
Poster panel: 336

Poster Number:

Measurement of Dynamic Operating Range of Discovery MI Digital PET/CT for O-15 Myocardial Blood Flow Imaging with an Anthropomorphic Torso Phantom (#1698)

J. Teuho1, 2, R. Siekkinen1, 3, T. Tolvanen1, 2, V. Saunavaara1, 2, 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: We investigated the dynamic operating range of the Discovery MI digital PET/CT system (GE Healthcare) for quantitative imaging with 15O-water in a high counting rate phantom study. The operating range of the system in terms of maximum counting rate was determined.
Materials and Methods: We performed a phantom study with an antrophomorphic torso phantom filled with scattering material and a high amount of O-15. PET imaging was performed with a test-retest protocol. The phantom imaging consisted of O-15 bolus in a saline bottle (measured volumes: 381 ml and 377 ml) inserted in the heart cavity of the phantom. The injected doses for both studies were 1525 MBq and 1537 MBq at scan start time. The count rate curves, dead-time correction factors (DTFs) and scatter fractions (%SF) were extracted and uptake in the saline bottle was measured.
Results: Normal detector operation was assumed with prompt count rates of 12.69 Mcps and 12.74 Mcps. The above counting rates resulted to DFT < 1.7 (60 % of detector live-time) with scatter fractions 32.95 % and 34.28 %.
Discussion: The dynamic operating range of O-15 H2O perfusion imaging on the Discovery MI digital PET/CT system was determined. The system shows improved counting rate and dead-time performance over previous imaging systems of the same vendor.
Conclusions: Based on the results achieved with the torso phantom, O-15 imaging should be performed below the maximum prompts rate of 12.69 Mcps.

Keywords: dynamic range, cardiac PET, O-15
Poster panel: 339

Poster Number:

Material Decomposition Radiography with Multilayer Detectors (#1793)

D. W. Kim1, 2, J. Kim1, H. Jeon2, Y. Ki2, S. Ha1, 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, Republic of Korea
3 Pusan National University, Center for Advanced Medical Engineering Research, Busan, Republic of Korea


Dual-energy (DE) x-ray radiography can enhance lesion conspicuity by reducing or removing the anatomical background clutter. Combining with the material decomposition technique, the dual-energy radiography may perform the lesion characterization. In this study, we have developed a multilayer detector for the capability of DE imaging at a single x-ray exposure. The multilayer detector is realized by stacking two phosphor-coupled photodiode arrays. We extend the multilayer detector technique to include material decomposition. Aluminum (Al) and PMMA are chosen for two known basis materials. For the calibration, we have fabricated the Al and PMMA in wedge forms for the consideration of their various thicknesses at a time. The projection signals for the wedge phantoms are described by the polynomials of two material thicknesses from the calibration. Pixel signals obtained for an arbitrary object can then be expressed as the linear combination of two material thicknesses by solving the inverse of the polynomials numerically. We have validated the linearity of material-specific pixel signals using the wedge phantoms. We will verify the applicability of material decomposition radiography using the multilayer detector to bone and soft-tissue images of a mouse-mimetic phantom. The results will be discussed in comparisons with those obtained from the conventional DE method.

Keywords: Dual-energy imaging; Sandwich detector; Material decomposition; Radiography;
Poster panel: 342

Poster Number:

Timing calibration for low-biased TOF PET reconstruction with clinical dataset (#1888)

W. Zhu1, S. Tang2, Y. Lv2, Y. Dong2, H. Li1

1 UIH America, Inc., Houston, Texas, United States of America
2 United Imaging Healthcare, Shanghai, China


Detectors in time-of-flight PET system may be subject to timing drifts over time, which result in biased TOF reconstructed PET images. The scope of this paper is to propose a technique which could continuously use clinical patient data for timing calibration in TOF PET systems.

Assuming non-TOF PET data suffers only trivially from timing drifts, a non-TOF reconstruction may be firstly performed to obtain a “bias-free” PET image. The image is then projected into data space with the desired TOF kernel, corrected with necessary calibrations (estimated randoms and scatters, etc), to obtain the estimated TOF prompt data. Following that, a block pair-wise fitting between the estimated TOF data and the collected one is done to measure the drift for each block pair. Finally the anti-drift shift may be applied to correct the TOF data for final image reconstruction.

In the phantom study, we deliberately altered the timing offset table in the system and reconstructed the cylindrical phantom with non-TOF and TOF reconstructions. The non-uniformity score was 1.3% for non-TOF, 4.2% for TOF before calibration, and 1.6% for TOF after. With clinical datasets, the TOF images also resulted in less relative bias after calibration (e.g. in the example they were 5.7% (before) and 2.1% (after) in liver ROI, 5.0% (before) and 1.9% (after) in heart ROI), compared with non-TOF PET image.

The proposed method could use patient PET/CT data to drive the timing calibration for scanners continuously. The calibrated timing information could maximize the performance of TOF PET systems.

Poster panel: 345

Poster Number:

Application of Conditional Adversarial Networks for Automatic Generation of MR-based Attenuation Map in PET/MR (#1980)

L. Tao1, X. Li2, J. Fisher1, C. S. Levin1

1 Stanford University, Radiology, Stanford, California, United States of America
2 University of Arizona, Center for Gamma Ray Imaging, College of Optical Sciences, Tucson, Arizona, United States of America

This work is partially supported by GE Healthcare.


Current PET/MR imaging systems use methods based on MR image segmentation with subsequent assignment of empirical attenuation coefficients for attenuation correction in PET image reconstruction. Delineation of bone in MR images has been challenging, especially in the head and neck areas, due to the difficulty of separating bone from air. In this work, we study deep learning techniques to automatically generate attenuation maps directly from MR images, with focus on the head and neck areas. We use a generative adversarial network (GAN) in a conditional setting for this image translation task. GANs separate the deep learning network into a generator, which tries to generate fake examples, and a discriminator, which learns to distinguish between real and fake examples, and train the two networks simultaneously. The objective function of the conditional GAN is a combination of the generator loss, discriminator loss, and the L1 distance between the label image and the output image from the generator. Image pairs of PET/MR image (input) and corresponding PET/CT based 511 keV photon attenuation map (label) are used to train the network. The network is trained for 6k iterations. The generator loss is trained from 2.0 to 1.4. The discriminator loss is trained from 0.6 to 1.0. The L1 loss is trained from 0.2 to 0.1. In our previous work with a basic autoencoder network for the conversion from MR images to corresponding attenuation maps, if we convert the defined L2 loss for the network to the RMS pixel value prediction error, the autoencoder network, after training, has a pixel prediction error of around 0.2 when all pixel values are scaled from 0 to 1. In this study, the L1 loss also represents this pixel prediction error, which is around 0.1 after the network is trained. This indicates an average reduction of 50% of the pixel prediction error.

Poster panel: 348

Poster Number:

An Accurate Time Calibration Method with Phantom Position Correction for EXPLORER Scanner (#2124)

X. Lu1, S. An1, W. Liu1

1 Shanghai United Imaging Healthcare Co., Ltd, Shanghai, China


As the industry partner of the EXPLORER consortium, we have designed and built a high-resolution, high-sensitivity PET/CT scanner which is expected to have high performance for total body imaging. The device has a ring diameter of 786 mm and an axial field of view of 1940 mm. The device consists of 8 sections, each section consists of 24 modules. The detector modules are composed of arrays of LYSO crystals of dimensions 2.76x2.76x18.1 mm. Read-out is performed using SensL J-series SiPMs. From the actual acquisition of PET detector signals to the final image reconstruction, due to the unique time delay in each channel, a wide coincidence time window is necessary for true coincidence events. In order to achieve the best timing performance, the time offset must be aligned. Furthermore, accurate time alignment is critical for TOF image reconstruction. In this work, we introduce an weighted linear least squares approach which can robustly estimate the time offsets using data from all possible detector pairs, of which errors can be well controlled. A cylinder phantom filled with FDG or Ge68 solution located in the FOV. Here, the phantom position can be off centered FOV, which make phantom positioning easily. All possible valid events from the phantom are collected. The phantom position is used as a correction factor in this method. For each LOR, there exists a time difference spectra, calculating the centroid time difference, variance of the spectra and total event number. The reciprocal of variance divided by total event number is used as the weight factor later.  Using this approach, a time alignment accuracy of better than 10ps can be achieved. Depending on the system stability, time offsets should be aligned regularly. It does not require extra hardware. Furthermore, this approach provides a meaningful result within a reasonable computing time.

Keywords: Total body PET, Time alignment, EXPLORE, off center correction
Poster panel: 351

Poster Number:

Calibration of PET Detectors Based on Monolithic Blocks Using Voronoi Diagrams (#2212)

M. Freire2, 1, A. González-Montoro1, A. J. González1

1 Universitat Politècnica de València, Instituto de Instrumentación para Imagen Molecular (I3M), Valencia, Spain
2 Universidad de Valencia, Burjassot, Spain


In this work, we show a method to calibrate PET detector modules composed by monolithic scintillator blocks. The method is based on Voronoi diagrams. Those permit to calibrate, in only one iteration, the energy and the 3D photon impact coordinates. We have tested the method using an array of collimated radioactive sources.

The calibration procedure was carried out by acquiring flood maps of an 11x11 22Na sources. After the acquisition, the Voronoi diagrams are calculated using MATLAB, in total we have 121 different Voronoi cells. Then, the calibration factors for the impact coordinates are calculated. In the case of the energy calibration, the photopeak channels of each cell are referred to the central one. Finally, we calibrate the data points doing an interpolation based on the inverse square distance of the data point and its closer Voronoi cells. We are currently working on the calibration of the depth of interaction coordinate too.

To evaluate the performance of this method, we used a PET detector with a 50x50x15 mm3monolithic LYSO coupled to an array of 12x12 SiPM (3x3mm2elements). We have compared the results with a calibration based on a 1D polynomial approach. The results regarding XY positions were successful. On one hand the show very good linearity and, on the other hand, they allow one to calibrate the 100% of the crystal volume. The last advantage is not achieved when using for instance the 1D polynomial approach. Moreover, we have determined about 10% improvement in the spatial resolution with the suggested approach. Regarding energy resolution, the results were also good, allowing correcting for light scintillation truncations (light losses) especially at the edges.

We will detail the procedure and accurately described the methodology and results for a variety of crystal thickness and treatments. We are currently working on the depth of interaction calibration also using this method.

Keywords: Monolithic crystals, detector calibration, SiPM
Poster panel: 354

Poster Number:

Implementation and Validation of Image Reconstruction for PET Data From GE SIGNA PET/MR Scanners In the STIR Library (#2300)

P. Wadhwa1, 2, K. Thielemans3, O. Bertolli3, N. Efthimiou4, E. Emond3, B. A. Thomas3, M. Tohme5, G. Delso6, W. Hallett2, R. Gunn2, D. Buckley1, C. Tsoumpas1, 2

1 University of Leeds, Department of Biomedical Imaging Science, Leeds, United Kingdom
2 Invicro Ltd., LLC, London, United Kingdom
3 University College London, Institute of Nuclear Medicine, London, United Kingdom
4 University of Hull, School of Life Sciences, Hull, United Kingdom
5 GE Healthcare, Waukesha, United States of America
6 GE Healthcare, Cambridge, United Kingdom


Software for Tomographic Image Reconstruction (STIR: http://stir.sf.net) is an open source C++ library for reconstruction of emission tomography data. This work describes and validates recent developments made in STIR to read and reconstruct the PET acquisition data extracted from GE SIGNA PET/MR scanners.

Data handled include list-mode and normalisation files (both in Hierarchical Data Format (HDF5) file format) and Magnetic Resonance Attenuation Correction (MRAC) images and associated files for attenuation of the bed and coils. We describe the modifications to STIR that allow accurate histogramming of this list-mode data in the same sinogram-organisation as the scanner, and the implementation of the data corrections for the GE SIGNA PET/MR. This allows reconstruction of acquisition data with all data corrections using STIR, independent of any software supplied by the manufacturer.

The implementation was validated by comparing results of the various data corrections and final image reconstruction by comparing with the vendor-supplied toolbox. The reading of data from HDF5 files shows a small difference of 0.05% in histogrammed counts and an overall difference of 1.05% for randoms correction sinogram. Images were reconstructed using STIR and the GE toolbox using the ordered subset expectation maximisation (OSEM) reconstruction algorithm. The STIR reconstructed images have similar resolution and quantification but have some remaining artefacts due to geometric, deadtime corrections and alignment between the PET and MR gantries that will be addressed in future work.

This work will enable the usage of all current and future STIR algorithms, including penalised image reconstruction, motion correction and direct parametric image estimation, on data from GE SIGNA PET/MR scanners. This allows the development of novel algorithms and evaluation on data from a clinical PET/MR scanner in a fully open source framework.

Keywords: positron emission tomography, pet-mr, Image Reconstruction, Data Corrections
Poster panel: 357

Poster Number:

Motion Compensation for Free-Breathing Diffusion-Weighted Imaging (MoCo DWI) in Whole Body Integrated PET-MRI (#1773)

C. Dávid1, 3, T. Vahle2, R. Grimm2, B. Kiefer2, P. Bachert1, 3, M. Kachelrieß1, 4

1 German Cancer Research Center (DKFZ), Division of Medical Physics in Radiology, Heidelberg, Baden-Württemberg, Germany
2 Siemens Healthcare GmbH, Erlangen, Bavaria, Germany
3 Heidelberg University, Faculty of Physics and Astronomy, Heidelberg, Baden-Württemberg, Germany
4 Heidelberg University, Medizinische Fakultät Heidelberg, Heidelberg, Baden-Württemberg, Germany


Diffusion-weighted imaging (DWI) is often used in PET-MRI whole-body acquisitions. Due to the required acquisition time of several minutes, respiratory motion is a major concern in the abdomen. Current techniques to solve the introduced motion blurring artifacts are breath-hold commands or (prospective) gating. The acquisition time implies repeated breath-hold phases, while gating prolongs the measurement about twofold — both are unfavorable for clinical use. A third technique is motion compensation (MoCo) which allows free-breathing image acquisitions. For this the respiratory motion is estimated using deformable image registration. However, image-based registration is challenging on DWI because these images have decreasing signal-to-noise ratio with increasing b-value. To overcome this limitation we propose the use of a separate golden-angle, stack-of-stars spoiled gradient-echo (GRE) scan. This scan is already used and available for PET/MR MoCo, thus the acquisition time is not extended. We use the joint-MoCo-HDTV algorithm for the motion estimation, and apply the motion model to compensate respiration during the DWI exam. Then the warped images are averaged. Volunteers are scanned with the previously described GRE followed by a single-shot echo-planar DWI pulse sequence. We compare the result with reconstructions without motion compensation (noMoCo) and retrospective gating. The noMoCo images suffer from significant motion blurring artifacts and the gated images show low SNR. In comparison to that our proposed MoCo DWI has similar noise level as the noMoCo images but similar sharpness to the gated images. The proposed approach promises improved ability to detect small lesions in DWI and more accurate apparent-diffusion-coefficient (ADC) maps.

Keywords: pet-mr, motion compensation, diffusion-weighted imaging, respiratory motion
Poster panel: 360

Poster Number:

Joint Correction of Attenuation and Scatter Using Convolutional Neural Networks (CNN) for Time-of-Flight PET (#2705)

J. Yang1, D. Park2, Z. J. Wang1, Y. Seo1

1 UCSF, Department of Radiology and Biomedical Imaging, Physics Research Laboratory , San Francisco, California, United States of America
2 Microsoft, Bellevue, Washington, United States of America


Purpose: This study aims to develop and optimize a convolutional neural network (CNN) for joint correction of attenuation and scatter in time-of-flight (TOF)-enabled PET/MR imaging, directly from non-attenuation/scatter-corrected PET (PETNASC) images as well as accelerating attenuation/scatter corrections for PET reconstruction.
Methods and Materials:
CNN architecture: The proposed CNN is based on the U-Net but modified for joint attenuation/scatter correction (JASC). The CNN consists of encoder and decoder layers with concatenated connections between corresponding encoder-decoder layers. At each layer, 2D convolution is performed with 3 × 3 kernels, batch normalization, and rectified linear unit (ReLU), followed by 2 × 2 downsampling (max pooling) for the encoder; and 2D deconvolution is performed after upsampling (bilinear interpolation) for the decoder. This model was implemented using Tensorflow libraries.
Ratio Net: Our proposed architecture is ratio learning (RatioNet) that learns the ratio of attenuation/scatter-corrected PET (PETASC) and PETNASC for more effective and efficient training, instead of learning PETASC directly. This idea comes from residual learning (ResNet) that learns the difference between inputs and outputs by adding a residual connection from input to output, though standard U-Net learns to generate output directly.
Data set: Pancreas-focused 18F-FDG scans were performed for 45 minutes in 22 subjects in a TOF-enabled PET/MR (SIGNA, GE Healthcare) scanner. We utilize paired PETNASC and PETASC images to train the proposed model that can predict PETASC (output) from PETNASC (input) images.
Discussion: Organs and their boundaries (e.g., lung, cardiac, bone, skin, etc) are perceptible in PETASC, PETNASC, and even the ratio map of PETASC and PETNASC. For this reason, CNN can extract patterns from PETASC and PETNASC, and our proposed model can predict PETASC from PETNASC through training our proposed RatioNet.

Keywords: PET, Convolutional Neural Network, Attenuation Correction, Scatter Correction, Deep Learning
Poster panel: 363

Poster Number:

Improving PET parametric image quality using kernel-based image denoising and gradient-free curve fitting algorithm (#1066)

H. - M. Huang1

1 National Taiwan University, Institute of Medical Device and Imaging, Taipei City, Taiwan


Parametric images obtained from kinetic modeling of dynamic PET provide a new way of visualizing quantitative parameters of the tracer kinetics. However, due to the finite amount of data and image noise, parametric images often suffer from poor image quality. The aim of this study is to improve the quality of parametric images. We propose a kernel-based image denoising method for dynamic PET images. We also evaluate whether the gradient-free curve fitting algorithms can provide improved parametric image quality. Computer simulation was performed to evaluate the performance of the proposed method. The simulation results show that the proposed method can produce better Ki image than the maximum likelihood expectation maximization (EM) algorithm. In addition, the Ki image obtained from the proposed method was comparable to that obtained from the kernel-based EM reconstruction. Finally, we show that the gradient-free curve fitting algorithms can further improve the quality of parametric images.

Keywords: image denoising, PET parametric imaging
Poster panel: 366

Poster Number:

Direct 4D Patlak reconstruction in dynamic FDG PET imaging with population-based input function (#2056)

Q. Ye1, 2, Z. Lyu1, 2, S. Yao3, Y. Dong4, W. Zhu5, H. Liu6, J. Wu7, Y. Liu1, 2, T. Ma1, 2

1 Tsinghua University, Department of Engineering Physics, Beijing, China
2 Ministry of Education (Tsinghua University), Key Laboratory of Particle & Radiation Imaging, Beijing, China
3 Chinese PLA General Hospital, Beijing, China
4 Shanghai United Imaging Healthcare, Shanghai, China
5 UIH America Inc., Houston, Texas, United States of America
6 Yale University, Department of Internal Medicine (Cardiology), New Haven, Connecticut, United States of America
7 Yale University, Department of Radiology and Biomedical Imaging, New Haven, Connecticut, United States of America


Introduction: In dynamic PET, direct parametric reconstruction is advantageous over conventional kinetic modeling method in terms of reduced impact from noise. Clinical application of dynamic PET is limited by technical complication in measuring the input function from patient blood samples. Population-based input function (PBIF) is an alternative approach and has been demonstrated to be feasible in regions of interest (ROI) -based kinetic analysis. The objective of this work is to investigate the performance of direct 4D Patlak reconstruction in dynamic FDG PET imaging with PBIF.

Methods: A direct 4D Patlak reconstruction strategy with PBIF was proposed. Poisson noise model and nested technique were used. Dynamic images and unscaled parametric images were generated from direct Patlak reconstruction with unscaled PBIF. The scaling factors estimated from dynamic images were used after the direct reconstruction due to the linear relationship between dynamic images and kinetic parameters. Phantom simulation, preclinical scans and ongoing clinical scans were performed to compare the indirect and direct reconstruction with PBIF.

Results: The Ki images generated from direct analysis were less noisy than those generated from indirect analysis. In phantom studies, the directly reconstructed images at convergence shows less noise and comparable bias. For both simulated and experimental data, similar scaled input functions were obtained in the indirect and direct analysis with PBIF. With shorter acquisition time, more iterations are required to reach convergence, while the advantage of direct reconstruction is more remarkable.

Conclusion: Our proposed direct reconstruction strategy with PBIF was demonstrated to be feasible. Compared with Ki images generated from indirect analysis, directly reconstructed Ki images showed better performance in terms of less noise and comparable bias.

Keywords: FDG PET, direct reconstruction, population-based input function
Poster panel: 369

Poster Number:

4D Reconstruction Improves Robustness of SPECT MBF to Dose Reduction (#2699)

R. G. Wells1, J. M. Renaud1, R. A. deKemp1, T. D. Ruddy1

1 University of Ottawa Heart Institute, Cardiac Imaging, Ottawa, Ontario, Canada


Dedicated cardiac SPECT cameras can be used to measure myocardial blood flow (MBF) and flow reserve (MFR) but lower dose studies would be preferred. We evaluate 4D image reconstruction using temporal B-splines for reducing the impact of activity reduction on 99mTc-SPECT MBF estimates.

SPECT and PET MBF studies were acquired in 25 patients (BMI=27.4 +/- 3.7 kg/m2).  PET MBF was measured with standard clinical protocols.  SPECT data were acquired on a cardiac pinhole camera in listmode using a rest (303 +/- 70 MBq) / stress (1110 +/- 168 MBq) 1-day 99mTc--tetrofosmin protocol.  Data were subsampled to create data sets with ¼ of the original counts.  For full- and ¼-activity data, reconstruction was done 1) independently for each time frame using MAP-EM (3D) and 2) for all frames simultaneously using a 4D MAP-EM algorithm (4D) based on temporal B-spline basis functions. MBF was calculated using a 1-tissue-compartment model and a previously determined extraction fraction correction. The correlation between SPECT and PET MBF with full-count studies was compared to that with ¼-counts for both 3D and 4D reconstruction.

For 3D, the PET to SPECT correlation in global MBF was r=0.88 (full count) and r=0.80 (¼-count). For 4D, the correlations were slightly lower at r=0.85 and r=0.76 for full and ¼-count data respectively. For the higher-count stress data, the change in correlation between full and ¼-count data was similar for 3D and 4D: r=0.79 vs r=0.73 (3D) and r=0.74 vs r=0.65 (4D).  For the lower-count rest datasets, the change in correlation was larger for 3D (r=0.49 vs r=0.20) than for 4D (r=0.54 vs r=0.38). The change in MFR (stress MBF / rest MBF) was also larger for 3D (r=0.65 vs r=0.46) than for 4D (r=0.52 vs r=0.47).

4D image reconstruction may better preserve the accuracy of SPECT MBF calculations, compared to 3D, when injected activities are reduced. However, the lower absolute correlations with PET indicate that further work is needed to improve overall performance.

Keywords: SPECT, Myocardial Blood Flow, Cardiac, 4D reconstruction
Poster panel: 372

Poster Number:

Multi-color imaging of GNP in phantoms towards theranostic imaging of tumours (#2903)

A. Castoldi1, 2, C. Guazzoni1, 2, K. Ricketts3

1 Politecnico di Milano, DEIB, Milano, Italy
2 Istituto Nazionale di Fisica Nucleare, Sez. Milano, Milano, Italy
3 University College London, Research Department of Tissue and Energy, Division of Surgery and Interventional Science, London, United Kingdom


The results presented in this contribution stem out a collaborative interdisciplinary effort aimed at identifying the most suitable imaging techniques to map the distribution of functionalized nanoparticles acting as biological markers taken up by specific cellular receptors in healthy and pathological biological tissues. Nanoparticles can be labelled to hypoxia seeking agents or other tumour bio-parameters; X-rays can then be used to probe the markers at depth, with contrast given by the nanoparticles. The uptake of nanoparticles by cancer cells could also enable detection of small clusters of infiltrating cancer cells which are currently missed by commonly used imaging modalities.

We have already demonstrated in previous works the feasibility of use of a L-XRF system in measuring gold NP concentrations and this work will focus on the development and validation of a K-XRF detection system in order to ease penetration through overlying tissue. The K-absorption edges of Gold, Gadolinium and Silver are conveniently located (80.7 keV, 50.2 keV and 25.5 keV, respectively) so we developed a custom compact detection module which accommodates a CdTe pn diode detector (4 × 4 mm active area) and all the needed biasing and frontend electronics. The detection module is conceived to allow the possibility of coupling mechanical collimators or X-ray poly-capillary lenses for high energy to perform 2D imaging. The contribution will focus on the advantages of the use of different types of NPs in cancer imaging and on the results of the experimental qualification carried out with different NPs (Gold, Gadolinium and Silver) and samples to assess the technique detection limit, linearity and sensitivity as a function of the NP material. Interesting images of relevant phantoms will be presented.

Keywords: gold nanoparticles, XRF, biomarker, theranostic, tumour imaging
Poster panel: 375

Poster Number:

A comparative study of AML, NEGML and OSEM based on experimental and clinical 90Y-PET data using the CASToR platform (#1395)

M. Millardet1, S. Moussaoui2, D. Mateus2, J. Idier2, M. Conti3, T. Carlier1

1 INSERM UMR 1232, CRCINA, Nantes, France
2 CNRS UMR 6004, LS2N, Nantes, France
3 Siemens Medical Solutions USA, Inc., Knoxville, Tennessee, United States of America


Yttrium-90 is an isotope used in nuclear medicine for the radioembolization of hepatic tumours. This β--emitter also produces positrons with a branching ratio of 3.2 × 10−5 which has paved the way of PET imaging as a post-therapy microspheres distribution assessment. PET imaging in this context is challenging because of the very small amount of data collected, and the positive bias in low activity regions introduced by classical reconstruction algorithms such as OSEM. This issue can be tackled by using algorithms allowing negative values in the image space such as AML and NEGML. However, to our knowledge, a full assessment of these algorithms using experimental and clinical data was still missing. For that purpose, the CASToR software was used. Its unified core allows to objectively compare different algorithms. We studied AML and NEGML with their different parameters based on four sets of data with different numbers of prompt coincidences. It was found that the parameter of NEGML did not affect significantly quantitative results. The images derived using this algorithm were bias-free in low-activity regions but the noise is highly increased in those areas. Depending on the parameter chosen, AML went continuously from OSEM to an algorithm close to NEGML. For a well-chosen value of this parameter (which depended on the activity level) AML reduced the positive bias while keeping a noise level similar to OSEM. This reduction ranged between 10 % for a relatively high number of prompt coincidences (around 6 millions, random fraction: 76 %) to 60 % for low number of prompt coincidences (around 2 millions, random fraction: 90 %). We also verified that the measured activity remains correct in every regions of the image.

Poster panel: 378

Poster Number:

About Measurement of PET Spatial Resolution (#1569)

M. A. Lodge1, J. Sunderland2, A. Rahmim1

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


Introduction: Spatial resolution is an important parameter for assessing the performance of positron emission tomography (PET) systems but this apparently simple measurement is complicated by various issues including the reconstruction algorithm, parameter settings and acquisition methodology. Here we describe a series of experiments that aim to untangle some of these issues.
Methods: Measurements of PET spatial resolution were made using 3 different experimental arrangements: (1) a point source in air; (2) a point source in a radioactive background; and (3) a uniform cylinder positioned at a slightly oblique angle, allowing spatial resolution to be measured from the phantom edge response function. All experiments involved 18F and were performed on a clinical scanner, Biograph mCT.
Results: Ordered-subsets expectation maximization including time-of-flight (OSEM+TOF, no point spread function modelling) rapidly converged to a stable full-width at half-maximum (FWHM) for all 3 geometries. OSEM+TOF provided better spatial resolution than Fourier rebinning plus filtered back-projection (FORE+FBP). FWHM degraded with increasing radial distance but in a complex way, apparently dependent on the reconstruction algorithm. FWHM derived from point sources and OSEM+TOF were similar with or without a radioactive background (radial FWHM at 10 cm, 4.77 ± 0.11 mm vs. 4.78 ± 0.06 mm). A uniform cylinder positioned at an oblique angle allowed measurements of radial FWHM (4.74 ± 0.17 mm) that were very consistent with point source measurements.
Conclusion: Spatial resolution derived from iterative reconstruction (without point spread function modeling) rapidly converged to a stable FWHM and was very consistent for 3 different experimental arrangements. These results support use of the uniform cylinder phantom as a tool for resolution assessment that may be particularly useful for multi-center evaluations due to its simple set-up.

Keywords: cylinder, performance assessment, PET, point source, spatial resolution
Poster panel: 381

Poster Number:

Improved lesion detectability and short imaging times with long axial field-of-view TOF PET (#1903)

S. Surti1, V. Viswanath2, M. E. Daube-Witherpsoon1, J. S. Karp1

1 University of Pennsylvania, Department of Radiology, Philadelphia, Pennsylvania, United States of America
2 University of Pennsylvania, Department of Bioengineering, Philadelphia, Pennsylvania, United States of America


Current generation whole-body, time-of-flight (TOF) PET scanners typically have a 16-25 cm axial field-of-view (AFOV) and have a coincidence timing resolution in the range of 300-600ps. In recent years there has also been an interest in developing whole-body PET scanners with much longer AFOV that not only increase the system sensitivity but also allow imaging more (if not all) of the patient in a single bed position. One important benefit of very high sensitivity is the potential to significantly reduce routine clinical scan times which can be beneficial in reducing patient motion artifacts or allowing for respiratory gating. As part of the EXPLORER consortium we have completed construction of the 70-cm long PennPET Explorer scanner with 250ps timing resolution. The aim of this study is to study the impact of a long AFOV PET scanner in reducing the total scan time using lesion detectability as the metric. Previous detectability studies using Monte Carlo simulations demonstrated the benefit of longer AFOV (72 cm vs 18 cm) and improved timing resolution (300ps vs 600ps), indicating a potential factor of ten reduction in total scan time with the longer AFOV and improved timing resolution. These simulation studies will guide the measurements that will be performed with the PennPET Explorer scanner. We will perform lesion detectability studies on this scanner using data acquired with a realistic anthropomorphic Alderson phantom and embedded lesions. Generalized scan statistics methodology will be used to measure the area under localized receiver operating characteristic (ALROC) curves as a function of scan duration. The results of this study will subsequently be used to help develop imaging protocols for human imaging with the PennPET Explorer scanner.

Keywords: TOF PET, total-body PET, lesion detectability, scan time
Poster panel: 384

Poster Number:

Evaluation of the Administrated Dose and Image Quality of Pediatric Patients (#2159)

H. Kertész1, T. Traub-Weidinger2, I. Rausch1, T. Beyer1, J. Cal-Gonzalez1

1 Medical University of Vienna, QIMP group, Center for Medical Physics and Biomedical Engineering, Vienna, Wien, Austria
2 Medical University of Vienna, Division of Nuclear Medicine, Department of Biomedical Imaging and Image-Guided Therapy, Vienna, Wien, Austria


The aim of this study is to evaluate PET image quality and the effective dose to pediatric patients following 18F-FDG PET/CT imaging when employing different image reconstruction methods. 

Data from 10 pediatric patients with epilepsy and 5 oncological patients was evaluated. All patients underwent an 18F-FDG examination on a Siemens TPTV PET/CT system. The acquired list mode data was preprocessed to reduce the number of counts to 75%, 50%, 35%, 20% and 10% of the total counts. The data was reconstructed using the STIR framework. Three reconstruction methods were evaluated in all cases: OSEM with Gaussian filter (FWHM = 0, 2, 4 mm), Maximum A Posteriori (MAP) algorithm with median root prior (MRP) (β = 50, 100 and 200) and MAP with quadratic prior (QP) (β = 10, 50 and 100). Lesion-to-background ratios (LBR) were calculated for hot lesions, while the LBR between the epileptic focus and the corresponding contra-lateral side of the brain was calculated for hypo-metabolic regions. The uniformity within a background region was also evaluated. Relative deviations of the LBR at reduced number of counts were calculated by comparison with the values obtained with all the counts. The patient effective dose (ED) from the PET and low-dose CT was also evaluated. 

Relative deviations below 10% were obtained when counts are reduced to 50%. The relative deviation in the LBR was: OSEM (4mm-filter) - 7.0%,  MRP (β = 200) - 6.7%, QP (β = 100) - 21.2%. The noise level for these cases was 15.4%, 17.6% and 27.8%, respectively (14.3%, 14.6%, 21.2% with all the counts). Using 50% of counts the total ED was reduced up to 53% and 68% of the standard ED, respectively. 

In conclusion, a potential reduction of the injected tracer amount and subsequent patient exposure from the PET examination of up to 50% can be achieved without compromising the PET image quality by applying a suitable image reconstruction method.

Poster panel: 387

Poster Number:

Towards a Performance Measurement Protocol for Positron Emission Mammography Systems (#2546)

L. F. Torres-Urzua1, H. Alva-Sánchez1, A. Martínez-Dávalos1, O. Garcia-Perez2, E. Barrera-Garcia2, M. Rodríguez-Villafuerte1

1 Instituto de Fisica, UNAM, Fisica Experimental, Mexico City, Mexico
2 Instituto Nacional de Cancerologia, Mexico City, Mexico


In this work we propose two phantoms and a set of measurements specifically aimed at evaluating positron emission mammography (PEM) systems. The lack of international protocols for the evaluation of PEM systems led us to design image quality phantoms to quantify spatial resolution, uniformity, lesion detectability, recovery coefficients (RC) and spill-over ratios (SOR). The first phantom consists of a cylindrical PMMA container in which a set of interchangeable inserts can be placed and filled with a positron emitter. One insert is a microDerenzo phantom with a series of hot-rod cylindrical hollows (1-5 mm diameter) arranged in a pie-like distribution to assess the system resolution. Another insert consists of hollow microspheres with internal diameters between 4.9 and 10 mm to simulate hot/cold lesions immersed in a warm/cold background to assess lesion detectability. A third insert, inspired by the NEMA NU 4-2008 image quality phantom for small-animal PET scanners, has two sectors: one to obtain the RCs as a function of hot cylinder diameter and another to calculate the SORs from cold cylinders filled with water and air. Finally, this phantom can be used to obtain a uniform volumetric activity distribution if no insert is used. The second phantom, consisting of a thin PMMA container, in which as many as 5 thin capillary tubes mimicking line-sources can be inserted, was developed to measure the line spread function (LSF). With the proposed phantoms we evaluated the performance of the Naviscan PEM Flex Solo II, a commercial PEM system commonly used in the clinic, using 18F and 68Ga to include the impact that positron range has on the reconstructed images. Our results indicate, for example, that positron range from the energetic positrons from 68Ga (radial LSF FWHM=4.22±0.13 mm and tangential LSF=4.59±0.29 mm) has a considerable detrimental effect over the spatial resolution compared to 18F (radial LSF FWHM=2.92±0.10 mm and tangential LSF=3.02 ± 0.15 mm).

Poster panel: 390

Poster Number:

Attributing Uncertainties in the Identification of Hotspots in SPECT Imaging (#2533)

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

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


In SPECT imaging, the identification of hotspots relies either on visual inspection of the reconstructed
tomographic images or on post-processing image analysis methods. Both approaches do not provide the
capability to attribute a quantifiable uncertainty to this identification. We present a scheme which allows
for the quantification of this uncertainty, based on the "Reconstructed Image from Simulations Ensemble"
(RISE) method. RISE results in the reconstruction of the image and the measurement of the uncertainty
associated with it. Parameters defining the uptake of radioactivity, the position and the size of a hotspot,
and as well as their associated uncertainties, are derived from the projection data. The capabilities of the
proposed scheme are demonstrated with data from numerical phantom simulations.

Keywords: SPECT, Image Reconstruction, Error Analysis, Uncertainty
Poster panel: 393

Poster Number:

Comparison of synthetic CT generation methods for MRI-only radiation planning in the pelvic region (#1064)

H. Arabi1, J. A. Dowling2, N. Burgos3, X. Han4, P. B. Greer5, N. Koutsouvelis6, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Switzerland
2 CSIRO, Australian e-Health Research Centre, Queensland, Australia
3 Inria, Aramis project-team, Paris, France
4 Elekta Inc., Maryland Heights, United States of America
5 University of Newcastle, Department of Physics, New South Wales, Australia
6 Geneva University Hospital, Division of Radiation Oncology, Geneva, Switzerland


Accurate radiation dose calculation is a major challenge in MRI-only radiation treatment planning as the required electron density map is not readily provided by this modality. In this work, a selection of promising novel synthetic CT generation methods, recently reported in the literature, were evaluated using the same cohort of patients and evaluation metrics. This includes four atlas-based approaches, namely median value of atlas images (ALMedian) (Sjölund et al., Phys Med Biol 2015), atlas-based local weighted voting (ALWV) (Dowling et al., Int J Radiat Oncol Biol Phys 2015), bone enhanced atlas-based local weighted voting (ALWV-Bone) (Arabi et al., Phys Med Biol 2016), iterative atlas-based local weighted voting (ALWV-Iter) (Burgos et al., Phys Med Biol 2017), a machine learning approach based on deep convolutional neural network (DCNN) (Han, Med Phys 2017) and a segmentation technique into a single tissue class (water-only). Organ auto-contouring from MR images was performed for bladder, rectum and bones where ALWV, ALWV-Bone, ALWV-Iter and DCNN exhibited comparable performance while DCNN showed slightly higher segmentation accuracy resulting in Dice indices of 0.93±0.17, 0.90±0.04, and 0.93±0.02, respectively. The dosimetric evaluation demonstrated that ALMedian, ALWV, ALWV-Bone, ALWV-Iter and DCNN led to similar mean dose errors within organs at risk and target volumes showing less than 1% dose discrepancy. The two-dimensional gamma analysis performed at 1%/1 mm criterion demonstrated comparable pass rates of 94.99±5.15%, 94.59±5.65%, 93.68±5.53% and 93.10±5.99% for ALWV-Bone, DCNN, ALMedian and ALWV-Iter, respectively, whereas ALWV and water-only resulted in pass rates of 86.91±13.50% and 80.77±12.10%, respectively. DCNN and advanced atlas-based methods exhibited promising dosimetric and segmentation accuracy (DCNN is slightly better) suggesting that the challenge of synthetic-CT estimation from MR images can be resolved with a clinically tolerable error.

Keywords: MRI-only radiotherapy planning, CT synthesis, Treatment planning;, PET/MRI, dose calculation
Poster panel: 396

Poster Number:

A Study on Boron-10 Concentration Estimation in Boron Neutron Capture Therapyusing a Current-mode Energy-resolving “transXend” Detector  (#1255)

I. Kanno1, D. Nishimatsu1

1 Kyoto University, Nuclear Engineering, Kyoto, Japan


This paper shows the results of feasibility study on boron-10 (B-10) concentration estimation in boron neutron capture therapy (BNCT). One of the problems in BNCT is the difficulty of B-10 concentration measurement in cancer tissues. For the measurement, prompt gamma rays with 0.478 MeV emitted by Li-7 atoms can be used. Gamma ray background, however, is very high in the BNCT facility, due to 2.2 MeV gamma rays emitted by the reactions of H(n, ?)D, and accompanied annihilation 0.511 MeV gamma rays. In a previous study, a thick collimator was placed in front of a gamma ray detector for avoiding background gamma rays. In such a configuration, imaging of B-10 concentration was very difficult. With the purpose of imaging B-10 concentration, a current-mode “transXend” detector is employed in this feasibility study. The transXend detector consisted of five 5×5×10 mm3 TlBr segmented detectors measured gamma rays as electric current, and gave gamma ray energy distribution after analysis. A 5 cm diameter cancer tissue with various B-10 concentrations from 10 to 50 ppm was placed in a 18 cm diameter 20 cm length water phantom. Gamma rays created by neutron reactions were calculated by a simulation code PHITS. By using an unfolding code SAND-II, gamma ray energy spectrum was estimated in 223 energy bins with 10 keV energy ranges. For the estimation of B-10 concentration using the numbers of gamma rays with 0.478 and 2.2 MeV in energy, three-layered neural network was employed. The obtained B-10 concentration showed good agreement with the expected ones.

Keywords: boron neutron capture therapy, B-10 concentration
Poster panel: 399

Poster Number:

Development of Single Crystal CVD Diamond Membrane Microdosimeters for Particle-therapy (#1288)

M. T. Pomorski1, P. Barberet2, 5, P. Bergonzo11, J. Herault10, W. Kada7, L. de Marzi4, T. Pourcher8, 9, D. Tromson3, S. Saada1, N. Skukan6, I. Zahradnik1

1 CEA-LIST, Diamond Sensors Laboratory, Gif sur Yvette, France
2 CNRS, UMR5797, CENBG, Gradignan, France
3 CEA-LIST, Sensors and Electronic Architectures Laboratory, Gif sur Yvette, France
4 Institut Curie, Centre de Protonthérapie d'Orsay, Orsay, France
5 Université de Bordeaux, CENBG, Gradignan, France
6 Ruder Boskovic Institute, Division of Experimental Physics, Zagreb, Croatia
7 Gunma University, Faculty of Science and Technology, Gunma, Japan
8 CEA, Laboratoire TIRO, Nice, France
9 Université de Nice-Sophia Antipolis, Nice, France
10 Centre Antoine-LACASSAGNE, Institut Méditerranéen de Protonthérapie, Nice, France
11 CEA-LIST, Gif sur Yvette, France

On behalf of DIAMIDOS-Collaboration


Radiotherapy, especially hadron therapy (with protons and carbon ions), is an innovative type of cancer treatment. By using hadron beams, tumor cells can be more effectively destroyed than by using conventional photons. At present, knowledge of the linear energy transfer (LET) is a key parameter in predicting the relative biological effectiveness (RBE) of such particle beams. However, over recent years limitations of using LET as a predictor for the RBE has received increasing attention. Under the consideration of such limitations, measureable stochastic quantities such as  the lineal energy has been introduced to provide a fundamental basis for microdosimetry. This is essential for validating simulations and models currently in use and for improving treatment efficiency. Today, no such microdosimetric detector is available on the market for routine clinical use. The existing alternative technologies are tissue equivalent proportional counters (TEPC) and solid-state silicon based microdosimeters. However, these devices are yet to fully fulfil all severe requirements for radiation quality assurance in hadron therapy. CEA-LIST-Diamond sensors Laboratory (LCD) is developing a new tissue equivalent diamond membrane based microdosimeter, along with the necessary electronics for such, to obtain precise measurements of the lineal energy distributions in clinical conditions with high rates and high spatial resolution. Interdisciplinary experimental techniques, such as nano- and microfabrication, as well as device characterization under ion nuclear microprobes, have been employed for diamond micro-dosimeter prototyping. The first test was to evaluate the performance of devices’ prototypes with clinical hadron beams, providing the basis for them to be later implemented in radiobiological numerical models.

Keywords: microdosimetry, diamond
Poster panel: 402

Poster Number:

Development of Real-time Range Monitor for Particle Therapy using Scintillation Fiber Imager (#1502)

H. Numakura1, F. Tokanai2, K. Morimoto3, T. Iwai2, S. Kurosawa2, 4, T. Umebayashi2, Y. Ohashi1, T. Moriya2, M. Murata5

1 Yamagata University, Graduate School of Science and Engineering, Yamagata, Japan
2 Yamagata University, Research Institute, Yamagata, Japan
3 RIKEN, Nishina Center for Accelerator-Based Science, Wako, Japan
4 Tohoku University, New Industry Creation Hatchery Center (NICHe), Sendai, Japan
5 the University of Tokyo, Department of Computational Diagnostic Radiology and Preventive Medicine, the University of Tokyo, Tokyo, Japan


Particle therapy is globally growing cancer treatment, whose good dose conformity is a main advantage. In particle therapy, range uncertainty is one of the main subject to be solved. The actual delivered dose distribution, particularly the Bragg peak depth, is influenced by anatomical changes inside the body. These changes might lead to an underdose at the tumor or an overdose at surrounding normal tissues. Although it is ideal that the actual Bragg peak depth is monitored during irradiation, such real-time monitoring method has not yet been established. During therapy involving carbon ion irradiation, a substantial number of secondary particles are created along the path of the carbon ions via nuclear reactions including nuclear fragmentation. Because some of them have enough energy to escape from the body, they can be utilized as a source to monitor the Bragg peak depth. The position where the escaped secondary particles are created can be estimated by a tracking detector system equipped outside the body. We have been developing a scintillation fiber imager (SFI) for use as a particle-tracking detector. The SFI consists of two sets of double layers of scintillation fibers. The layers are aligned perpendicular to each other to enable two-dimensional imaging. Each layer consists of 64 square fibers with a cross section of 0.25 mm2. Eight multianode photomultiplier tubes (Hamamatsu R5900-L16) are used to read out the signal of the SFI. The imaging capability of the SFI was investigated by using an 80 MeV proton beam at Tohoku University Cyclotron and Radioisotope Center in Japan. It was confirmed that a fine 2D image of the proton beam can be obtained. The detection efficiency was measured by comparing the count rate of the SFI with that of a conventional plastic scintillation detector. The uniformity of the sensitivity, the position resolution, and the detection efficiency of SFI will be presented.

Keywords: carbon ion therapy, scintillation fiber imager, range uncertainty, secondary particle detection
Poster panel: 405

Poster Number:

First results with MACACO II: second prototype of a Compton telescope (#1589)

A. Ros García1, L. Barrientos1, J. Barrio1, J. Bernabéu1, M. Borja Lloret1, P. Dendooven3, A. Etxebeste1, L. Gabarda1, M. C. Jimenez Ramos2, C. Lacasta1, R. Marco1, E. Muñoz1, J. F. Oliver1, I. Ozoemelam3, J. Praena2, J. Roser1, C. Solaz1, G. Llosá1

1 IFIC, CSIC–University of Valencia, E-46071 Valencia, Spain, IFIMED (IFIC), Valencia, Spain
2 Centro Nacional de Aceleradores, Universidad de Sevilla , Sevilla, Spain
3 KVI-CART, University of Groningen, KVI-CART, Groningen, Netherlands


The IRIS group at IFIC Valencia is developing a three-layer Compton camera for treatment monitoring in ion beam therapy. The system is composed of three detector planes, each made of a LaBr3 monolithic crystal coupled to SiPM arrays. The first prototype (MACACO) was fully characterised in the laboratory and in beam tests demonstrating the feasibility of the proposed technology. A second prototype (MACACO II), described in this work, is currently being developed to improve performance. The SiPM arrays have been replaced by newer models, leading to an improved detector energy resolution which translates into a higher spatial resolution of the telescope. A new coincidence board allows acquisition of data with any two or all three detectors simultaneously. In addition, the image reconstruction code has been improved with an accurate model of the sensitivity matrix. The device has also been tested in two accelerator facilities. The first involved a proton beam of 18 MeV impinging a graphite target. The results allowed the successful reconstruction of two target positions separated by 5 mm. The second test beam used protons of 150 MeV impinging on a PMMA target. Preliminary analysis results show prompt-gamma energy spectra as expected for these proton energies. Analysis of this data is ongoing, with the aim of reconstructing the Bragg peak profilein the PMMA target. Further improvements to the readout electronics are also underway.

Keywords: hadron theraphy, test beam, compton imaging, SiPM, continuous crystals
Poster panel: 408

Poster Number:

Annihilation Gamma Imaging for Carbon Ion Beam Range Monitoring using Si/CdTe Compton Camera (#1653)

R. K. Parajuli1, M. Sakai1, W. Kada2, K. Torikai1, M. Kikuchi1, K. Arakawa1, M. Torikoshi1, T. Nakano1

1 Gunma University, Gunma University Heavy Ion Medical Center, Maebashi, Japan
2 Gunma University, Graduate School of Science and Technology, Kiryu, Japan


Carbon ion radiotherapy is an advanced cancer treatment technology because of its strong impact on tumor due to maximum energy deposition at the end of their range in target. Reducing the range ambiguities could sophisticate the current advanced carbon ion radiotherapy treatment leading to better conformity and consequently improved treatment. One of the methods could be the on-beam annihilation gamma imaging. In this study, Silicon/CadmiumTelluride detector based Compton camera, ASTROCAM 7000HS developed by Mitsubishi Heavy Industries Ltd., was implemented for carbon beam monitoring. Carbon ion beams generated in Gunma University Heavy Ion Medical Center were used for the beam monitoring on polymethyl methacrylate phantom. We were able to reconstruct the image using the energy events of 511 keV annihilation gammas emitted due to the nuclear reaction of carbon ion beams of 290 MeV/u with the polymethyl methacrylate phantom. Moreover, the effect of the beam range shift of 30 mm, 60 mm and 90 mm in the beam delivery were clearly visualized in the reconstructed images of 511 keV at 25 mm, 52 mm and 76 mm respectively. The experimentally measured values were consistent with the shift in Bragg peaks (by 25.5 mm, 51.3 mm and 76.70 mm) evaluated using Monte Carlo simulation (PHITS 2.820) performed using water equivalent range shifter (30 mm, 60 mm and 90 mm respectively).

Keywords: Carbon ion radiotherapy, Compton camera, Beam Monitoring, Annihilation gamma
Poster panel: 411

Poster Number:

Maximum-likelihood reconstruction and simulation analysis on Compton imaging of astatine-211 for targeted α-particle radiotherapy (#1703)

Y. Nagao1, 2, M. Yamaguchi1, S. Watanabe1, N. S. Ishioka1, N. Kawachi1, H. Watabe2, 1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Takasaki Advanced Radiation Research Institute, Takasaki, Japan
2 Tohoku University, Cyclotron and Radioisotope Center (CYRIC), Sendai, Japan


Astatine-211 is a promising radionuclide for targeted α-particle radiotherapy of cancers. It is required to image the distribution of targeted radiotherapeutic agents in a patient’s body before or during treatment for optimization of treatment strategies and determination of the suitability of a given agent for a particular patient. Because the biodistribution of 211At is different from that of 131I, which is a common radiohalogen in conventional planar or single-photon emission computed tomography imaging, it is important to image 211At directly. The 211At and its daughter radionuclide of 211Po emit gamma rays (570, 687, and 898 keV) at the total intensity of 0.9%. Recently, we have proposed to image 211At with the gamma rays using a developed Compton camera, and demonstrated the imaging capability of the camera in an experiment of a point-like 211At source by backprojection algorithm. In this study, we implemented list-mode maximum-likelihood expectation-maximization (MLEM) algorithmto the experimental data to improve the image. Moreover, Monte Carlo simulations were performed using the Particle and Heavy Ion Transport code System (PHITS) to analyze the experimental results. As a result, the position resolution andbackground noise were improved in the MLEM image compared with those in the back-projected image in both the experiment and simulation. Future research plans include the sophistication of the simulation analyses.

Keywords: Astatine-211, Compton camera
Poster panel: 414

Poster Number:

Assessment of Neutron Dose for Pediatric Patient According to the Proton Beam Delivery Technique (#1778)

D. - H. Kim1, Y. Han1, B. Lee1, E. Shin1, S. Park1, S. Cho1, D. H. Lim1

1 Samsung Medical Center, Sungkyunkwan University School of Medicine, Department of Radiation Oncology, Seoul, Republic of Korea


The neutron dose varies depending on the delivery technique of proton therapy. The purpose of this study was to evaluate the neutron dose equivalent produced in proton therapy for the pediatric patient and to evaluate changes in neutrons dose equivalent with treatment delivery techniques. We retrospectively selected 42 pediatric patients who have treated by wobbling proton therapy. We investigated maximum proton energy, ridge filter, field size, snout position, and air gap that were used in the wobbling plan and 4 scenarios were generated by combining the most used items. For each scenario, the wobbling, line scanning, and line scanning with aperture plan were generated using the treatment planning system. All of the scenarios were implemented using the GEANT4 toolkit for evaluating the neutron dose. The ambient neutron dose equivalents were calculated by the neutron energy spectra measured in the 12 cm diameter receptor. The receptors were placed at the distance of r = 0, 25, 50, 100, 150, and 200 cm from the isocenter. The ambient neutron dose equivalent of line scanning beam was 90.9%, 89.3%, 92.5% and 94.7% smaller than that of wobbling beam, respectively. And the ambient neutron dose equivalent of line scanning with aperture beam was 91.3%, 86.2%, 85.3%, and 92.9% smaller than that of wobbling beam, respectively. In proton therapy, the wobbling beam is used because the lateral penumbra has a dose benefit compared to line scanning beam, but the exposure by the neutron dose is increased. However, using line scanning with aperture beam not only reduces the lateral penumbra but also the neutron dose is not as high as wobbling beam. In the present study, only the external neutron dose was evaluated. However, future studies will investigate the effect on the individual organs of the pediatric patient, including internally scattered neutron dose in the human body.

Keywords: Ambient neutron dose equivalent, Wobbling proton therapy, line scanning with aperture proton therapy, Monte Carlo simulation
Poster panel: 417

Poster Number:

A prototype scintillating fibre based beam profile monitor for hadron therapy beams  (#1848)

B. D. Leverington1, M. Dziewiecki1

1 Universitaet Heidelberg, Physikalisches Insitut, Heidelberg, Baden-Württemberg, Germany


A preliminary prototype detector for online monitoring of the position and size of a hadron therapy beam is presented as a possible replacement of the current MWPC-based tracking chambers of the Heidelberger Ionenstrahlungs-Therapiezentrum (HIT) at the end of their life. The gas-based tracking chambers have a large ion drift time (150 μs) and are the limiting factor for the application rate of dose through raster-scanning at HIT.

The proposed device is based on a thin ribbon of scintillating fibres of 250 μm diameter and readout with 64-ch photodiode arrays with 0.8 mm channel pitch. It has the benefit of a fast scintillation decay time (<7 ns), high beam rate capability, and smaller channel granularity. Five and two-layer fibre mats were tested as proof-of-concept and light versions of the detector plane and are foreseen to be made even lighter by removing epoxy. Monte Carlo simulation has shown that 30% more material than currently present would increase the lateral beam spread by only 2 to 8%.

Fibre planes based on SCSF-3HF and SCSF-78MJ fibres from Kuraray were characterized at HIT. The average position resolution for single 95 μs integration frames for ion beams measured at the isocenter was found to be better than 100 μm, but strongly depends on the beam spot size and intensity. An average beam width reconstruction resolution better than 350 μm was found. The raw-data readout rate is 10 kHz. These results exceed the performance specifications specified by the clinic for a replacement detector. In addition to performance studies, aging studies due to radiation damage from regular accelerator activities are ongoing. The expected annual dose at a central beam spot is approximately 1 MGy. The 3HF fibres are observed to perform better overall and meet the required beam reconstruction performance specifications of HIT. As such, further development is ongoing to meet the additional requirements of a low mass detector along with improved performance characteristics.

Keywords: scintillating fibres, hadron therapy, beam instrumentation, photodiodes, scintillator
Poster panel: 420

Poster Number:

Compton-PET imaging of 10C for Range Verification of Carbon Ion Therapy (#2114)

A. Mohammadi1, E. Yoshida1, Y. Okumura1, M. Nitta1, F. Nishikido1, A. Kitagawa1, 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 Maximilian University, Department of Medical Physics, Munich, Germany


In proton and carbon ion therapies, visualization of the range of incident particles in a patient body is important for treatment verification. In-beam positron emission tomography (PET) imaging in ion therapy is one of the method for verification of treatment due to the high quality of PET images. We have already shown the feasibility of radioactive beams of 15O and 11C using in-beam PET imaging using our OpenPET system. Recently, we have developed a whole gamma imaging (WGI) system which can work as PET, single gamma-ray and triple gamma-ray imaging simultaneously. The WGI system has high potential to detect location of 10C, which emits positron with a simultaneous gamma-ray of 718 keV, and activity of other produced positron emitting nuclei within patient body during ion therapy. In this work, we focus on investigation of performance of WGI system for 10C irradiation. First, the performance of scatterer detector of the WGI system regarding the radiation hardness was studied for 10C irradiation, then the performance of WGI was studied by simulation using the Geant4 code. The scatterer detector consisted of a segmented GAGG crystal and a multi-pixel photon counter (MPPC) and its performance was studied as a PMMA phantom was irradiated with 5 and 400 spills of 10C with energy of 350 MeV/u. No damage to the MPPC (scatterer detector) was observed even though after irradiation of 400 spills. Sensitivity values of WGI system by simulation of a 10C nuclide at the center of field of view (FOV) for PET mode imaging, single gamma-ray of 718 keV imaging and triple gamma-ray imaging were 7.85%, 0.28% and 0.012% respectively. The performance of WGI system is going to be evaluated for 10C irradiation and the results will be discussed in the conference.

Keywords: Compton PET, Ion therapy, Range verification
Poster panel: 423

Poster Number:

Dosimetry Calculation in Human Glioblastoma for Radiotherapy: a Graphical User Interface with Monte Carlo simulations (#2348)

F. A. A. Slimani1, M. Hamdi2, V. Hubert-Tremblay1, P. Delage1, M. Bentourkia1

1 Université de Sherbrooke, Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada
2 Université de Mostaganem, Département de Génie Électrique, Mostaganem, Algeria


We developed a graphical user interface based on GEANT4 and GATE to calculate particle interactions and absorbed dose estimates in phantoms and small animals such as mice. In the present work, we extended the interface to absorbed dose calculation and particle interactions in humans based on 3D CT images and radiation beam definition in a clinical setting. The images of the subject and of the beams were supplied in DICOM format. All the parameters needed to calculate the absorbed dose were obtained from these DICOM images. In this work, images of a human brain with glioblastoma were used together with the parameters of five photon beams defined in the clinic. The DICOM file of the beams contained several parameter values such as beam energy, in this case 6 MV, the dose to be deposited in the tumor as a total of 60 Gy, and the 5 beams orientation. Based on the 3D CT images of the patient brain, the whole patient head was rebuilt from the voxel intensity and size providing the real dimensions and calculated density of the head. The results show how the primary photons and secondary particles interact in the brain, and a 3D dose grid similar in dimensions and spatial resolution to the supplied images of the brain was obtained representing the absorbed dose in Gy. In conclusion, with this new interface, it is simple to enter geometrical objects, animal or human 3D images, to select the appropriate tasks from menus, and to run the simulation without the need to be familiar with computer programming or investigating the many classes of GEANT4 or GATE. The interface can also be used to simulate any type of radiation beams provided by the DICOM-RT set of the subjects.

Keywords: Monte Carlo, Radiotherapy, Dosimetry, GUI, GEANT4
Poster panel: 426

Poster Number:

Personalized Particle Therapy with Simultaneous True Replica 3D Printed Organs and Digitized GEANT4 Simulations  (#2545)

C. Da Via1, J. Allison3, A. Aitkenhead2, A. Porter1, R. Price1

1 University of Manchester, Manchester, United Kingdom
2 NHS Christie Trust, Manchester, United Kingdom
3 Geant4 Associates International Ltd , Hebden Bridge, United Kingdom


In particle therapy a fast, accurate, and effective characterization of intensity and energy-modulated beams in 3-dimensions and patient-specific quality assurance of such beams is a considerable challenge. Each Intensity Modulates Proton Therapy (IMPT) beam in a treatment would normally use thousands of individual pencil beams of variable intensities, energies, and positions to produce highly inhomogeneous dose distributions within a target volume. The sum of contributions of all such beams corresponds to the desired dose distribution in the target.  For IMPT profile-scanning using conventional 2D detectors, normally ion chamber or semiconductor detectors, is not effective since the pencil beam sweeps continuously over a certain line in a given time, and any profile acquired with a moving detector would not be a representative of a true dose profile. Complex systems combining several layers of 2D detectors have been used with good results but limited spatial resolutions (~1cm) in the z direction where detector planes are usually spaced to accommodate readout electronics boards. To address this issue we propose the use of 3D-printed organs, which are a real replica, from the patient CT scan. The target volume within the replica is filled with liquid or plastic scintillating “voxels” to measure the true dose distribution in 3 dimensions on the target region and its surroundings. At the same time Geant4 Monte Carlo of the same organ and target region can be performed to simulate the 3D energy distribution so real data and simulations can be used in parallel to fully personalize the treatment planning. The paper will show results on both detection and simulations of simple 3D printed structures filled with liquid scintillator and readout with a SiPm after 20m long fiber with related simulations. It will also show results on the feasibility of the conversion software from CT-DICOM to STL formats and their successful import into a 3D printer and G4 .

Poster panel: 429

Poster Number:

A Portal Imaging System for Intensity Modulated Neutron Therapy Quality Assurance (#2592)

A. L. Lehnert1, A. González-Montoro1, 3, M. E. Kranz2, D. Q. DeWitt1, W. C. J. Hunter1, R. C. Emery2, D. C. Argento2, R. D. Stewart2, R. S. Miyaoka1

1 University of Washington, Radiology, Seattle, Washington, United States of America
2 University of Washington, Radiation Oncology, Seattle, Washington, United States of America
3 Instituto de Instrumentación para Imagen Molecular (I3M),, Valencia, Spain


The University of Washington (UW) Medical Center has undertaken the development of intensity modulated neutron therapy (IMNT) for the treatment of cancers resistant to low linear energy transfer radiations. This requires patient-specific quality assurance (QA), consisting of a comparison of simulated and measured 2D dose maps. However, no commercially available system exists to measure these maps.

We have designed a “portal imaging” system in which a piece of plastic is irradiated by the treatment field, creating positron‑emitting isotopes (C-11, O-15) through (n, 2n) reactions. It is then read in a nearby parallel-plate positron emission tomography (PET) system, creating a 2D map of neutron-induced activity and shape. An early prototype with smaller field of view (FOV) demonstrated linear response with dose and no saturation for a clinically relevant dose range.

A full-size imaging system is now being constructed with a 10 cm by 10 cm FOV utilizing repurposed PET detectors (Siemens ECAT EXACT HR+). Each of the two panels contains 4-by-4 detector modules, each with sixty-four crystals and four photomultiplier tubes. Data acquisition is done with UW Phase-II DAQ electronics with an appropriate integration time and custom software. Housing, as well as a fillable phantom for detector normalization, have been designed and are under construction.

All detector modules have been characterized (511-keV photon flood measurements), and all crystals are discernable in the crystal maps. Partial system (8 module) coincidence measurements showed an average spatial resolution of 3.25 mm FWHM, which improved to 2.5 mm when filtered to limit oblique lines of response. Application of an analytical normalization improved edge crystal spatial resolution by 5%. This performance is sufficient for verification of beam intensity and shape in future IMNT applications. Future testing will include imaging of the patient-specific IMNT treatment fields and comparisons to simulated activity maps.

Keywords: neutrons, neutron therapy, QC, PET, quality assurance
Poster panel: 432

Poster Number:

Detector Development for Prompt Gamma Imaging in Proton Therapy Online Monitoring (#2843)

H. Zhang1, 2, P. Fan1, 2, T. Xu1, 2, Q. Wei1, 2, Z. Wu1, 2, Y. Liu1, 2, T. Ma1, 2

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


Proton therapy allows for accurate positioning of target tumors by accurate dose delivery at the Bragg peak. However, a 10~15 mm proton range deflection could be caused by uncertainties including CT image artifacts and treatment setup errors. Prompt gamma imaging (PGI) offers a reliable real-time monitoring for proton range inside the patient and more accurate spatial information of the Bragg peak. In this work, we report the detector development for a prompt gamma imaging system. A detector based on a 12ⅹ12 BGO block and an 8ⅹ8 SiPM array was desigened. Since 2~8 MeV high energy of prompt gamma may cause complicated interaction inside scintillation crystal and saturation of SiPMs, positioning performance and response linearity of the detector was evaluated.

The proposed detector was tested under high energy sources, such as 22Na, 88Y and 232Th. SiPM signals were processed by a self-developed ASIC to obtain position signal X, Y and energy signal E which were then digitalized by an Analog-to-Digital convertor (ADC) on digital processing board (DPB). Flood maps at 511 keV and 1.275 MeV of 22Na, 898 keV and 1.836 MeV of 88Y and 2.6 MeV of 232Th were generated. Root mean squared (RMS) metric were defined to assess the detector positioning performance quantitively. Detector response linearity at different energy were evaluated.

The detector offered clearer flood map at higher energy, and RMS decreased when energy of incident photons increased. Energy resolution of 35.45%±6.62% (511 keV), 30.18%±4.66% (898 keV), 20.49%±3.20% (1.275 MeV) and 18.36%±4.21% (1.836 MeV) was reached respectively. Moreover, mean linearity of peak positions within 144 single crystals was 0.9998 which is a clear demonstration of good detector response linearity.

The result show that the proposed detector performed well on positioning, energy resolving and response linearity. Decreased RMS value and energy resolution implies even better positioning performance and energy resolving ability at 2~8 MeV.

Keywords: Proton therapy monitoring, Prompt gamma imaging, Scintillator detector, SiPM