2019 IEEE Nuclear Science Symposium and Medical Imaging Conference
supported by:

To search for a specific ID please enter the hash sign followed by the ID number (e.g. #123).

MIC Poster Session I

Session chair: Watson, Charles, C. (Siemens Medical Solutions USA, Inc, Molecular Imaging, Knoxville, USA); Daube-Witherspoon, Margaret, E. (University of Pennsylvania, Department of Radiology, Philadelphia, USA)
Shortcut: M-01
Date: Wednesday, 30 October, 2019, 1:40 PM
Room: Central 1
Session type: MIC Session


Click on an contribution to preview the abstract content.

Poster panel: 1

Poster Number:

Time-resolved simulation of optical modulation from ionization-induced fast charge carriers (#1009)

D. Jeong1, L. Tao1, J. Wang1, C. Levin1

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


To greatly advance coincidence time resolution for positron emission tomography (PET), one possible mechanism is to utilize the transient free carrier effect resulting from 511 keV photon interactions in a crystal. This paper studies the temporal characteristics of the prompt electron cascade produced at the early stage of ionization. It is challenging, however, to obtain precise timing information of carrier cascades, due to asynchronous decay of positrons. To address this problem, we propose to use relativistic electrons from ultrafast electron diffraction (UED) setup, which provides femtosecond (10-15 s) time resolution capabilities. We conducted simulation studies to understand the behavior of free carrier generation resulting from relativistic electrons, by expanding our previous work, which used 350 keV photo-electrons. Most UED facilities have higher electron energy than 511 keV, and we simulated 750 keV and 3 MeV for matching the actual experimental conditions. We report ionization electron cascade duration of 1.14 ps for 350 keV, 2.79 ps for 750 keV, and 12.1 ps for 3 MeV electrons in CdTe. We found that by decreasing the sample thickness, for example with 400 μm CdTe, the cascade time reduces to 2.1 ps, and absorbed the energy of 500 keV, approaching the regime relevant to PET. Time-resolved carrier cascade dynamics were simulated, which showed that the charge density decreased with higher energy (8.46x1018 cm-3 for 350 keV, 2.48x1018 cm-3 for 750 keV, and 1.02x1018 cm-3 for 3 MeV), due to charge cloud spreading over longer cascade times. Finally, we obtained detectable optical modulation strengths in the first 350um in the charge carrier trajectory in CdTe, with maximum intensity change of 3.05% over a 120 μm by 160 μm area in z-projection. These results indicate the feasibility of using relativistic electron simulations to understand the temporal characteristics of ionization-induced electron cascade events generated by 511 keV photons.

Keywords: Time-of-flight PET, Positron emission tomography, Ionization-induced optical-modulation
Poster panel: 4

Poster Number:

Two-Layer DOI Staggered GAGG Scatterer Detector for WGI (#1147)

S. Takyu1, E. Yoshida1, A. Mohammadi1, K. Kamada2, 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


We developed a prototype of the whole gamma imaging (WGI), in which a scatterer detector ring is inserted into a PET ring to add a Compton camera function. The scatterer detector ring consisted of 20 modules, each of which was made of 24 x 24 array of Gd3Al2Ga3O12(Ce) (GAGG) of 0.9 x 0.9 x 6 mm3 in size and a multi-pixel photon counter (MPPC) array (8 x 8, 3.2 mm pitch of 3 mm pixel, 50 micron subpixel size). However, its averaged energy resolution was 9.1 % at 662 keV, which was worse than potential performance of the material. We thought this was due to the degraded linearity of the MPPC signals. Therefore, using larger crystals to effectively spread scintillation photons inside crystal block was considered to be a solution. Also, improvement of the sensitivity without degrading spatial resolution was another issue. In this study, we developed a two-layer depth of interaction (DOI) staggered GAGG detector, based on the simulation results, which the WGI geometry with GAGG crystals 1.45 x 1.45 x 4.5 mm3 in size showed improved angular resolution measuring (ARM) and the sensitivity. The GAGG crystals were arranged to a 13 x 13 array for 1st layer and a 14 x 14 array for 2nd layer. The two-layer crystals were coupled to the same MPPC array with the resistor network. In the position map, almost all crystals were clearly separated. The two-layer detector showed better energy resolutions (7.8 % at 662 keV) than the first WGI prototype. 

Keywords: Compton camera, scatterer, DOI
Poster panel: 7

Poster Number:

TOF/DOI PET detector using a highly multiplexing readout based on striplines (#1273)

C. Tian2, H. K. Kim1, W. - S. Choong3, Y. Hua4, F. Xu2, J. Lyu2, Q. Xie2, C. - M. Kao1

1 University of Chicago, Department of Radiology, Chicago, Illinois, United States of America
2 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
3 Lawrence Berkeley National Laboratory, Berkeley, California, United States of America
4 Raycan Technology Co., Ltd, Suzhou, China


We propose to develop a TOF- and DOI-capable detector by using an array of pixelated LYSO crystals whose light outputs are read at both ends by SiPMs that match the crystals in size. A 2”x2” detector can contain 16x16 4mm crystals or 24x24 2mm crystals, yielding 512 or 1,152 SiPMs. Previously we have described a highly multiplexing readout method in which a number of SiPMs feed their outputs to a stripline (SL) at various defined locations. Analogous to TOF detection, the feeding location of a signal can be determined from the propagation differential time observed at the two ends of the SL. In this work, we use a single SL for all SiPMs in the front or rear array, therefore reducing the number of detector outputs to only 4. This equals 128x or 288x channel reduction. To our knowledge, this degree of channel reduction has not been seen in reported TOF detectors. The SL is designed so that the row (column) number of the active SiPM in the front (rear) array can be accurately estimated. Hence, the hit LYSO crystal is identified by the row and column numbers determined. DOI is calculated based on the relative amounts of light measured at the front and rear arrays, as described by others. In this work, we develop 4x4 arrays of 3.2 mm pitch to evaluate the proposed detector design. The LYSO size is 3x3x20 mm3 and the Hamamatsu MPPC arrays are used. The front and rear ends of the crystals are polished but the other surfaces are finished with 9 um roughness. With this surface treatment, we measure a DOI resolution of about 4 mm and an estimated CRT of 225.7 ps for a single crystal. For the 4x4 array, all crystals are clearly identified except for one (we postulate that this is caused by poor electrical contact of the SiPM with SL). The crystal-level pulse height spectra have well-defined photopeaks, showing an energy resolution in the range of 11.5-15.9%. At the meeting, full results including DOI resolution and CRT measured for the array, are reported.

Keywords: PET detector, multplexing readout, TOF, DOI
Poster panel: 10

Poster Number:

UTOFPET: a highly scalable TOF-PET detector concept (#1323)

N. Belcari1, 2, M. G. Bisogni1, 2, N. Camarlinghi1, 2, P. Carra1, 2, E. Ciarrocchi1, 2, G. Sportelli1, 2, M. Morrocchi1, 2, V. Rosso1, 2, M. D'Inzeo3, G. Franchi3, L. Perillo3, A. Puccini3, C. Bruschini4, E. Charbon4, F. Gramuglia4, E. Venialgo4, K. Deprez6, C. Thyssen5, R. Van Holen5, 6, M. Stockhoff5, E. Vansteenkiste6, S. Vandenberghe5

1 University of Pisa, Department of Physics , Pisa, Italy
2 INFN - Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, Pisa, Italy
3 AGE Scientific s.r.l., Capezzano Pianore (LU), Germany
4 Ecole Polytechnique Fédérale de Lausanne, Advanced Quantum Architecture (AQUA) Laboratory, Institute of Microengineering , Neuchâtel, Switzerland
5 Ghent University, Department of Electronics and Information Systems, Medical Image and Signal Processing (MEDISIP) , Gent, Belgium
6 Molecubes NV, Gent, Belgium


We present the design and preliminary simulated performance of a highly scalable TOF-PET detector concept. The UTOFPET project aims at the development of a TOF-PET detector module, with beyond-state-of-the-art performance, that is designed to be well suited for various applications, from brain-dedicated to total-body PET. The UTOFPET detector is based on a continuous scintillator crystal read out by 256 SiPMs, arranged in a 16 × 16 matrix whose outputs are processed by a stack of local processing and data acquisition boards. Initial simulations indicate the feasibility of on-board real-time position and time estimation with a spatial resolution below 0.5 mm FWHM for the whole detector.

Keywords: PET instrumentation, TOF-PET, SiPM
Poster panel: 13

Poster Number:

High-resolution PET detector with 100 ps coincidence time resolution using side-by-side phoswich design (#1338)

M. S. Lee1, J. W. Cates2, C. S. Levin1

1 Stanford University, Radiology / Molecular Imaging Program, Stanford, California, United States of America
2 Lawrence Berkeley National Laboratory, Applied Nuclear Physics Program, Berkeley, California, United States of America


With 100 ps full-width at half maximum (FWHM) coincidence time resolution (CTR), a 5-fold reconstructed image SNR gain can be achieved, assuming a 40 cm diameter subject, relative to no time-of-flight (TOF) capability. This excellent CTR can be achieved using a novel side-readout PET detector configuration, which allows higher light collection efficiency and lower photon transit time jitter. Here, we propose a new TOF PET detector for a high-resolution clinical system design that is referred to as “side-by-side phoswich detector” where two crystals with fast and slow decay time are side-coupled to the same photosensors.
GATE Monte Carlo simulation was conducted to find optimal condition for the best CTR for the side-by-side phoswich design. Based on the simulation results, a proof-of-concept PET detector was built that comprised two 1.9×1.9×10 mm3 LSO crystals (τd=34.39 and 43.07 ns) side-coupled to the same three 4×4 mm2 SiPMs. For timing analysis, the SiPM ‘fast output’ was fed into the amplifier and leading edge discrimination (LED) was applied for the time pickoff. For energy and crystal separation, the SiPM ‘standard output’ was fed into the scope and an in-house single time-over-threshold (ToT) circuit. All signal outputs were acquired with a digital oscilloscope and all coincidence measurements were repeated eight times with 8 distinct crystal pairs.
An average CTR of 107.07±3.17 ps was achieved at 32V bias voltage with a 2 mV LED threshold for the side-by-side phoswich configuration, with 94.46±1.78 ps and 107.99±2.26 ps CTR achieved for fast-fast and slow-slow single crystals, respectively. The side-by-side phoswich detector showed good global energy resolution of 10.35±1.20% at 511 keV and good crystal separation with the crystal separation accuracy of 0.95 for the two adjacent crystal elements when we applied ToT technique. This detector configuration will allow us to achieve high spatial resolution, ~10% energy resolution and 100-ps CTR at the same time.

Keywords: Time-of-flight, silicon photomultiplier, phoswich detector
Poster panel: 16

Poster Number:

Optimization of Crystal Surface Treatment and Reflector for SiPM-Based Dual-Ended Readout TOF-DOI PET Detector (#1391)

H. G. Kang1, T. Yamaya1, G. B. Ko2, J. S. Lee2, S. H. Song3, Y. B. Han3, S. J. Hong3, 4

1 National Institute of Radiological Sciences (NIRS) in National Institutes for Quantum and Radiological Science and Technology (QST), Department of Nuclear Medicine Science, Chiba, Japan
2 Seoul National University, Department of Nuclear Medicine, Seoul, Republic of Korea
3 Eulji University, Department of Senior Healthcare, Daejeon, Republic of Korea
4 Eulji University, Department of Radiological Science, Seongnam, Republic of Korea


In a previous study, we proposed a silicon photomultiplier-based (SiPM-based) dual-ended readout positron emission tomography (PET) detector that used a mean time method to achieve excellent time-of-flight (TOF) and depth-of-interaction (DOI) resolution simultaneously. However, the coincidence timing resolution (CTR) was greater than 300 ps since the crystal surface and reflector type were not optimized. The aim of this study was to investigate the optimal crystal surface treatment and reflector material to achieve a sub-200 ps CTR and sub-3 mm DOI resolution with a dual-ended readout PET detector using an LYSO crystal (2.9×2.9×20 mm3). The scintillation light inside the LYSO crystal was read out by two SiPMs (S11064-050P, Hamamatsu Photonics, Japan) using the dual-ended readout method. The CTR and DOI resolution were measured with two different crystal surfaces (polished and saw-cut) and three different reflector material scenarios of ESR (Air), ESR (Optical grease), and Teflon. We digitized the timing and energy signals by using a V775N TDC module (35 ps/bit) and V965 QDC module, respectively. The combination of the saw-cut LYSO crystal and ESR (Air) resulted in the best CTR (190 ± 4 ps) and DOI resolution (2.9 ± 0.2 mm) with the dual-ended readout configuration. We concluded the dual-ended readout method in combination with the saw-cut crystal and ESR (Air) reflector can provide a sub-200 ps CTR and sub-3.0 mm DOI resolution simultaneously.

Keywords: TOF, DOI, SiPM, PET
Poster panel: 19

Poster Number:

Material Budget Imaging - Feasibility study with small animal phantom (#1473)

H. Jansen1, P. Schuetze1

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


Tomographic methods for the imaging of complex structures are widely used in medical, industrial and scientific applications. Recently, we proposed a new imaging modality based on the measurement of the trajectory of electrons traversing a sample under test with electron momenta in the range of a few hundred MeV/$c$ up to several GeV/$c$. Each reconstructed electron trajectory yields the impact position on the sample and the scattering angle at the sample. Based on the measured impact positions the trajectories are assigned to image cells and the width of their scattering angle distribution is converted into a material budget value $\varepsilon$ of the traversed object at a given cell. The material budget $\varepsilon = x/X_{0}$ is the ratio of the path length in the material $x$ and the material's radiation length $X_{0}$. The contrast for a single projection in Material Budget Imaging (MBI) relates to the material budget traversed at a given image cell. As in CT, the 3D reconstruction of the material budget is realised by acquiring projections under different rotation angles and performing an inverse radon transform of the sinograms. Doses required for an acceptable image contrast at a desired voxel size are under study.
At the DESY II Test Beam Facility, MBI of a phantom of a rat's head was performed using an electron beam with momenta between 300 MeV/$c$ and 2 GeV/$c$ in combination with the DATURA Beam Telescope for high-precision particle tracking. We present a detailed study of this imaging modality comparing in-beam data at different particle momenta. The potential and limits of MBI are discussed in terms of the spatial resolution, the contrast-to-noise ratio and the precision of the absolute material budget reconstruction.

Keywords: Material Budget Imaging, Particle tracking, High energy electron imaging
Poster panel: 22

Poster Number:

Time-of-flight estimation in a PET block detector using convolutional neural networks (#1600)

E. Berg1, E. Mikhaylova1, J. Du1, S. R. Cherry1

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


Improved coincidence timing resolution (CTR) in time-of-flight (TOF) PET detectors can be achieved by using a non-linear statistical estimator instead of threshold methods such as leading edge and constant fraction discrimination. We previously demonstrated the use of convolutional neural networks (CNN) to estimate TOF from the pairs of coincidence digitized detector waveforms, achieved by training the CNN with a large number of waveform pairs each with known TOF. So far, these studies with CNNs have included only single crystal detectors. Here we investigate the use of CNN-TOF estimation in a PET block detector, related to what is used in the UIH uEXPLORER scanner, and investigate the use of multiple waveforms from each detector simultaneously as the input to the CNN. The detector uses four 6x6 mm SiPMs coupled at the corners of the LYSO array, and nine 3x3 mm SiPMs that fill the space between the corner SiPMs to increase light collection. Leading edge discrimination yielded CTRs of 451 ps and 387 ps using the four corner or all 13 SiPMs, respectively. Then, we trained 3-layer and 4-layer CNNs using events from all crystals and either the four corner SiPMs or all 13 SiPMs. The 4-layer CNNs achieved CTRs of 371 ps and 302 ps, using the four corner or all 13 SiPMs, respectively. As a preliminary investigation into the sensitivity of the CNNs to changes in the overall waveform shapes relative to the training dataset (e.g. drift in SiPM gain), we tested an extreme case by randomizing the rank order of the 13 SiPM waveforms before training, thereby losing all implicit spatial information in the data. This random-rank order trained CNN yielded a CTR of 351 ps with consistent-rank order test waveforms, and a CTR of 364 ps with random-rank order test waveforms. Although poorer than consistent-rank order training, the results indicate random rank-ordered training provides excellent model robustness, and may be a useful data augmentation strategy in the future.

Keywords: machine learning, convolutional neural network, time-of-flight, PET, signal processing
Poster panel: 25

Poster Number:

Towards the Development of a PET Detector based on TriMethyl Bismuth. Measurements of Free Ion Yield and Drift Charges in TetraMethyl Silane. (#1668)

B. Gerke1, S. Peters2, K. Bolwin1, V. Hannen2, C. Weinheimer2, K. P. Schäfers1

1 University of Münster, European Institute for Molecular Imaging, Münster, North Rhine-Westphalia, Germany
2 University of Münster, Institut für Kernphysik, Münster, North Rhine-Westphalia, Germany


Bold-PET is a new drift detector based on organometallic transparent dielectric liquid - TriMethyl Bismuth (TMBi). Due to its high atomic mass and density, TMBi offers a very high interaction probability for the photoelectric effect and is thus an excellent candidate to build PET detectors with high spatial resolution. In order to characterize organometallic liquids we have performed first tests with a similar but well-known liquid, TetraMethyl Silane. For charge transport evaluation we built a plate ionization chamber manufactured from electropolished ultra-high-vacuum compatible components. Ion yield and drifting charge measurements were performed in the test cell under the presence of a Cs-137 source. The results demonstrate a free ion yield in TMS of Gfi0=0.54 electron-ion-pairs per 100eV deposited gamma energy which is in accordance with previously measured values. Pulses generated by drifting charges could be quantified, indicating the capability to perform energy measurements.

Keywords: PET, detector, Positron Emission Tomography, TMBi, Trimethyl Bismuth
Poster panel: 28

Poster Number:

Development of a dual-ended readout detector for high-resolution PET (#1756)

M. Li1, S. Abbaszadeh1

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


Depth of interaction (DOI) information can improve the quality of reconstructed images of positron emission tomography (PET). DOI is especially important in high-resolution applications such as dedicated breast, brain and small animal imaging, where systems are compact and parallax error is notable. Among different DOI methods, the dual-ended readout can achieve great DOI resolution with better light collection efficiency. We designed a dual-ended readout detector for high-resolution PET. The detector was based on 4×4 lutetium-yttrium oxyorthosilicate (LYSO) units. Each unit contained 6×6 LYSO pixels, and the pixel size was 1×1×20 mm3. The four lateral surfaces of LYSO pixels were ground to W14 (roughness 10 to 14 μm) and the two ended surfaces were polished (roughness < 0.5 μm). The reflector was Toray Lumirror E60, which was an intermediate material between the specular reflector and diffuse reflector. Each LYSO unit was read out from both ends with two silicon photomultipliers. Another 1×25.8×20 mm3single LYSO slab was used to measure the designed detector performance at different depths. Data were acquired at 10 depths (1, 3, ..., 19 mm), and each depth had a 10 min acquisition time. The analog output signals from SiPM were digitized by PETsys TOFPET2 ASIC and acquired by PETsys SiPM Readout System. The ASIC and SiPM were cooled by a fan and a Peltier element. During the experiment, the SiPM temperature was controlled as 27.6±0.4 ◦C. The energy, time, and DOI resolution of the detector were characterized as 15.66%±0.66% at 511 keV, 602.98±10.58 ps and 1.40±0.05 mm, respectively.

Keywords: dual-ended readout, high resolution PET
Poster panel: 31

Poster Number:

Positron Emission Tomography with Sparse Block Rings and Continuous Bed Motion (#1801)

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

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


Clinical PET systems employ compact block rings to maximize sensitivity per axial field-of-view (AFOV) length, thereby resulting in considerable manufacturing costs. To reduce the cost per AFOV length, a sparse block rings configuration is modeled on a Siemens BiographTM mMR PET/MR scanner (sparse-mMR) by removing the coincidence counts from every other detector block ring. Moreover, Continuous Bed Motion (CBM) is performed along a limited distance to eliminate the axial sensitivity gaps. List-mode PET mMR data of the NEMA image quality phantom with 4:1 spheres-to-background ratio was acquired for 30min. The counts were binned assuming a sparse-mMR configuration and a CBM acquisition of constant speed along a distance of 2 blocks (16 detector rings, 6.4cm). The CBM scans were simulated by axially shifting the stationary mMR data to 16 different positions, including a reference, removing the counts associated with any even block rings, and shifting back to the reference position. The process was repeated with different input PET data of equal duration for all 16 positions and the corresponding output data were added. The CBM mode eliminated all axial sensitivity discontinuities by evenly acquiring data throughout the gaps. Contrast recovery (CR) and background variability (BV) were evaluated for PET images reconstructed from sparse-mMR CBM data against mMR and compact-½mMR stationary data of equal duration (~5min). The latter consisted of counts associated with only the 4 central mMR block rings. A similar CR performance was attained between the three configurations. Mean BV for sparse-mMR with CBM was 3.6% higher than mMR. Nevertheless, sparse-mMR with CBM attained a smoother axial variation in image noise, relative to compact-½mMR. CBM can restore continuity in the axial sensitivity profile of PET systems with sparse block rings to achieve similar contrast recovery and smoother noise variation, compared to compact systems, at half the cost or double the AFOV.

Keywords: Continuous Bed Motion, Sparse, Rings, PET, axial field-of-view
Poster panel: 34

Poster Number:

XEMIS2 Liquid Xenon Compton Camera for Small Animals 3γ Medical Imaging: Scintillation Light Measurements (#1876)

Y. Zhu1, S. Acounis1, N. Beaupère1, J. L. Beney1, J. Bert2, S. Bouvier1, C. Canot1, T. Carlier3, M. Cherel4, J. - P. Cussonneau1, S. Diglio1, D. Giovagnoli2, J. Idier5, F. Kraeber-Bodéré3, P. Le Ray1, F. Lefèvre1, J. Masbou1, E. Morteau1, J. S. Stutzmann1, D. Thers1, D. Visvikis2, Y. Xing1

1 SUBATECH, IMT Atlantique, CNRS/IN2P3, Université de Nantes, Nantes, France
2 INSERM, UMR1101, LaTIM, CHRU Morvan, Brest, France
3 Centre Hospitalier Universitaire de Nantes, Nantes, France
4 INSERM U892 équipe 13, Nantes, France
5 LS2N, Ecole Centrale de Nantes, CNRS/Inp, Université de Nantes , Nantes, France


An innovative Xenon Medical Imaging System, XEMIS2, designed for small animal 3g preclinical imaging, has been constructed and it is currently under test and qualification at the SUBATECH laboratory. It consists of a Compton camera, containing nearly 200 kg of liquid xenon, whose main goals are the precise three-dimensional localization of the 44Sc radioactive emitter used to image the small animal and the reduction of the administered radio-pharmaceutical activity in cancer diagnosis. The active volume of the XEMIS2 camera is surrounded by a set of PhotoMultiplier Tubes (PMTs) to measure scintillation light. The read-out anodes are segmented in 20000 pixels to measure ionization charges. In order to reduce the electronics dead-time during continuous data taking, a novel Data acquisition (DAQ) system specifically designed for XEMIS2 has been realized. It consists of two independent synchronized scintillation and ionization signal detection chains. The self-triggered scintillation light detection chain has been recently tested and calibrated in the prototype XEMIS1, whose experimental results showed a good performance. XEMIS2 will be soon installed at the Center for Applied Multimodal Imaging (CIMA) in the Nantes University Hospital for further preclinical studies. To safely manage a large amount of xenon in a hospital center, a recovery and storage cryogenic subsystem called ReStoX has been conceived, successfully commissioned and already installed at CIMA.

Keywords: 3 gamma imaging, Compton telescope, Functional imaging, Liquid xenon, Scintillation Light
Poster panel: 37

Poster Number:

A novel DOI detector design based on rectangular light sharing window technology (#1905)

B. Ye1, J. Yang1, S. Xie1, H. Peng1, Q. Huang2, J. Xu1, Q. Peng3

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


The depth of interaction (DOI) draws great attention due to its feasibility of alleviating the parallax error which deteriorated the image quality in peripheral FOV. A single-layer detector with rectangular light-sharing windows (RLSW) was designed to improve the accuracy of DOI decoding. The RLSW method replaces the reflector between segmented crystals with optical glue physically. The photons can propagate to surrounding crystals and the DOI information can be determined by analysis of signals.
With the internal radiation experiment and integration method, we analyzed the flood maps and defined the relation between position shifts in flood map and DOI information. The experimental results show: (1) The 9×9 LYSO crystals array measuring 2×2×20mm3 with 7mm RLSWs can be decoded clearly by the 6 mm SiPMs array. (2) The RLSWs show excellent performance of DOI decoding. (3) The mean absolute error (MAE) of the decoding accuracy in DOI is 1.9293 mm.
The results show light sharing window method is a promising method for accurate DOI decoding. The subject of our next research will focus on simulations included characterizing the surfaces treatment of crystals and size of windows. We plan to use the RLSW method for small animal positron emission tomography (PET) which meets preclinical and clinical applications.

Keywords: rectangular light-sharing windows, PET detector, DOI decoding, MAE resolution
Poster panel: 40

Poster Number:

Development of a Portable Beta Particle Detector for Precision Cancer Detection Using GEM (#1945)

D. Kim1, J. Yu1, Y. Chi1, M. Jin1

1 University of Texas at Arlington, Department of Physics, Arlington, Texas, United States of America


Radio-PTT, which combines radiotherapy and photo-thermal therapy (PTT), can achieve better local depletion of cancer cells and inhibition of tumor recurrence and metastasis than a stand-alone therapy. The radiolabeled photosensitisor can serve as an imaging reporter for image-guided light delivery. Gamma-rays are usually used for tomographic imaging, but heavy metal collimation limits its application on portable imaging devices. On the other hand, beta particles’ short range can be utilized for a compact design. In this study, we develop a Gas Electron Multiplier (GEM) detector for beta particle detection. The detector consists of a Kapton window, a drift foil, three GEM layers and a 128 channel anode readout board. The GEM detector, which amplifies weak signals by electron avalanches, is more sensitive to detect beta particles than gamma photons. By applying over 300 volts to each GEM layer, the detector can achieve over 104 effective gain. Beta emitter, 137Cs, is used to show that the constructed GEM detector can reliably detect beta particles. This study lays a foundation for further development of the portable GEM imaging detector through a scalable readout system, which will yield a high-resolution beta imaging tool for Radio-PTT.

Keywords: Radiotherapy, PTT, GEM, Beta particles
Poster panel: 43

Poster Number:

TOF Capabilities of Large Monolithic Blocks Combined with Analog SiPMs   (#2214)

E. Lamprou1, A. J. Gonzalez1, J. Barrio1, G. Cañizares1, M. Freire1, A. Gonzalez-Montoro1, L. Hernandez1, V. Ilisie1, F. Martos1, L. Vidal1, F. Sanchez1, J. M. Benlloch1

1 Institute for Instrumentation in Molecular Imaging, i3M, Valencia, Spain


Monolithic scintillators are good candidates for PET detector designs. They provide significant advantages in terms of accurate position decoding. In this contribution, we aim to provide a concluding report of the process towards extrapolating accurate timing resolution when combining monolithic blocks, analog SiPM photosensors and the TOFPET2 ASIC.  This work is a novel concept which may allow accurate 4D information of each gamma impact, avoiding the use of complex readout or detectors configurations.
We analyze the challenges shown up due to the wide scintillation light spread. A filtering method to discard falsie triggered events is presented, as well as the optimal timestamp assignment method is explored. A calibration procedure designed to correct the uncertainty introduced by time walk and time skew error is also described. CTR analysis at different DOI regions has been carried out. Experiments with single pixelated crystals are presented where a CTR below 200 ps FWHM has been achieved.  Eventually, a time resolution of 497 ps FWHM was reached for one detector module based on a large and thick LYSO monolithic block with 50x50x15 mm3 dimensions. This contribution is an incremental work showed in past meetings that comes now to a refined conclusion.

Keywords: TOF-PET, Monolithic crystal, SiPMs, ASIC.
Poster panel: 46

Poster Number:

A Low-Profile Positron Emission Tomography Front End for Submillimetric Resolution MRI Insert (#2387)

J. Bouchard1, 2, R. Espagnet1, 2, L. Arpin3, N. Moghadam1, 2, A. Samson1, 2, R. Lecomte4, 5, R. Fontaine1, 2

1 Université de Sherbrooke, Interdisciplinary Institute for Technological Innovation (3IT), Sherbrooke, Québec, Canada
2 Université de Sherbrooke, Department of Electrical and Computer Engineering, Sherbrooke, Québec, Canada
3 Imaging Research and Technology, Engineering, Sherbrooke, Québec, Canada
4 Université de Sherbrooke, Sherbrooke Molecular Imaging Center (CIMS), Sherbrooke, Québec, Canada
5 Université de Sherbrooke, Department of Nuclear Medecine and Radiobiology, Sherbrooke, Québec, Canada


The LabPET II, an avalanche photodiode-based submillimetric positron emission tomography (PET) scanner, offers great potential to integrate a magnetic resonance imaging (MRI) compatible insert owing to the inherently non-magnetic materials of the detectors and front-end electronics. Nevertheless, the overall geometry of the scanner is too large to fit inside the volume constrained by the bore and the radio-frequency antenna. This paper describes a new low-profile front end based on the LabPET II architecture dedicated to a mouse insert. It uses state-of-the-art 2.5D packaging and assembly processes such as wafer-level packaging, flip-chip assembly, ball grid array and ultra-compact surface mount components to integrate a high-density printed circuit board interposer detector. The interposer assembly is then attached to a carrier board which supports up to four interposers to achieve the required field-of-view for a mouse. Even with the substantially increased density of the electronics, the time-over-threshold energy resolution and the coincidence timing resolution show no signs of deterioration when compared to the standard LabPET II detection module.

Keywords: Positron Emission Tomography, 2.5D, PET insert, ultra-high density electronics, submillimetric
Poster panel: 49

Poster Number:

Performance of a Dual Layer PET Detector Using Time-Over-Threshold for Depth of Interaction with the TOFPET2 ASIC (#2484)

D. Prout1, M. Shustef1, A. Chatziioannou1

1 UCLA Crump Institute, Pharmacology, Los Angeles, California, United States of America


One approach to high sensitivity and high resolution PET is through phoswich block detectors consisting of multiple layers of scintillators. In our laboratory, we have developed phoswich detectors based on BGO and LYSO. These detectors provide both depth-of-interaction (DOI) information and cross-layer-crystal-scatter (CLCS) rejection through delayed-charge-integration (DCI) method. To achieve this, the DOI detectors require implementing extensive algorithms on a Virtex 6 FPGA connected to a bank of multiple 14-bit free running ADCs which sample the raw detector block signals at 125MHz. The high cost and poor scalability of such a data acquisition system limits the total number of detector channels that can be digitized. In turn, this demands heavy multiplexing of the signals and leads to complications at high event rates. The PETSys TOFPET2 ASIC is a cost effective and high performance chip designed for use with PET detectors. The main challenge with this ASIC is how to identify the crystal layers which provide the DOI information. Previously, we demonstrated that, in principle, the time, energy and time-over-threshold (ToT) provided enough information to discern which of the two layers an event had occurred. Here we demonstrate this in practice with a pair of recently constructed BGO/LYSO detector blocks using Hamamatsu MPPCs and the TOFPET2 ASIC. We present a method by which the ToT can be used to separate the events from the two layers and we also investigate the energy and coincidence timing performance of this detector. Results from this study show that ToT provided by the TOFPET2 ASIC is sufficient to identify in which crystal layer events occur. This method can also be used to identify and reject a large portion of the cross layer crystal scatter (CLCS) events. In addition, these preliminary results provide position, energy and timing for crystals in both the 19x19-crystal LYSO and 16x16-crystal BGO layers of this DOI detector.

Keywords: PET, DOI, BGO/LYSO, PETSys, Time over Threshold
Poster panel: 52

Poster Number:

Thermal Simulation of a 2.5D Stack-Up with Organic Interposer and Flip Chip for a Preliminary LabPET II Insert Design (#2535)

R. Espagnet1, J. Bouchard1, A. Lakhssassi2, R. Lecomte3, R. Fontaine1

1 Université de Sherbrooke, Interdisciplinary Institute for Technological Innovation 3IT, Sherbrooke, Québec, Canada
2 University of Quebec in Outaouais, Department of Computer Science and Engineering, Gatineau, Québec, Canada
3 Université de Sherbrooke, Sherbrooke Molecular Imaging Center (CIMS), Centre Hospitalier Universitaire de Sherbrooke, Department of Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada


The LabPET II is a positron emission tomography scanner developed for small animal imaging. The electronics include many printed circuit boards (PCBs) assembled with right angle connectors creating some areas hard to cool down. To alleviate this problem, a new electronic architecture is developed and consists of a Carrier Board (CB) on which four FR4-based PCB interposer are soldered in a flip chip technology. The aim of this work is to study a mouse-scanner field-of-view to evaluate different thermal management scenarii.
A prototype assembly of 12 CB creating a mouse scanner was developped with the thermal model of the LabPET II ASIC built in the FlowSimulation module of SolidWorks. The simulation evaluates three approaches: no thermal management, forced airflow (with a fan) and water-cooling. In the last case, three simulations were conducted to maintain the top face of the heat sink located on the carrier board at a temperature of 15, 20 or 30 degrees Celsius.
Without any thermal management, the ASIC temperature reaches 205 degrees Celsius. With an airflow, created with a fan directing an airflow axially outside the FoV, the ASIC temperature decreases below 100 degrees Celsius. With applied surface condition of 15 degrees Celsius on the heat sinks extremity parts, the ASIC maximum temperature is 65 degrees Celsius. When surface conditions of 20 and 30 degrees Celsius are located on the top of the heat sink, the ASIC temperature is reduced to 28 and 38 degrees Celsius respectively.
The new 2.5D architecture demonstrates that simple heat sinks can cool the temperature down to 25 degrees Celsius, a real improvement compared to the 40 degrees in the LabPET II architecture.

Keywords: LabPET II, Pet-insert, Thermal management, 2.5D assembly, Simulation
Poster panel: 55

Poster Number:

A Depth-encoding PET detector placing Horizontal-striped Glass on Crystal Arrays (#2609)

S. Han1, J. Kang1

1 Chonnam National University, Department of Biomedical Engineering, Yeosu, Republic of Korea


This study introduces a depth-encoding PET detector placing a horizontal-striped glass on the top of single scintillation crystals for providing continuous DOI information and reducing the number of output channel from N2 to N for N × N photosensor array. A Monte Carlo simulation was conducted to investigate the DOI capability for 4 × 4 LYSO-GAPD PET detector. Five different thicknesses of horizontal-striped glasses with 1 × 4 array ranged from 1 to 5 mm with 1 mm step. They were each placed on the top of crystal array. The simulation results showed no significant light loss was observed by employing light guide. We also estimated that the proposed estimation method for identifying interaction depth positions can provide accurate and uniform DOI resolution with increased dynamic range, compared to other conventional methods. The results of this study demonstrate that proposed DOI-PET detector with 3 mm-thick horizontal-striped glass could provide excellent 3D γ-ray interaction positions.

Keywords: Horizontal-striped glass, depth of interaction, light sharing, DOI-PET detector
Poster panel: 58

Poster Number:

Characterization of LYSO and CeBr3 Detectors with Lateral Sides Readout for a Multilayer Compton-PET (#2720)

J. Barrio1, N. Cucarella1, A. J. Gonzalez1, A. Gonzalez-Montoro1, V. Ilisie1, E. Lamprou1, F. Sanchez1, J. M. Benlloch1

1 Institute for Instrumentation in Molecular Imaging (i3M), Valencia, Spain


Two of the limitation factors of current PET scanners are their low sensitivity and lack of high image contrast. In order to improve these parameters, Compton interactions, which are usually discarded or unavoidably averaged, must be distinguished and included in the reconstruction process. To achieve this, we are working on two different approaches. In another contribution, a Compton-PET detector capable of distinguishing the Compton interactions that take place before the photoelectric effect within one monolithic crystal is being developed. In this contribution, a multi-layer detector approach is shown. We propose using thin scintillation layers with photodetectors coupled to all four lateral sides of the crystal. In the work presented here, a characterization of LYSO and CeBr3 detectors with lateral sides readout has been performed. The CeBr3 crystal also included a 5 mm thick quartz glass frame due to encapsulation constrains. The obtained results show an energy resolution of 12.1% and 9.1% FWHM for the LYSO and CeBr3 detectors, respectively. Pilot spatial resolutions ranging from 2.7 mm up to 4.3 mm for the LYSO crystal and from 4.1 mm up to 11.2 mm for the CeBr3 crystal were observed. These results suggest that the quartz glass frame surrounding the CeBr3 crystal significantly affects the light distribution of the optical photons reaching the photodetectors, worsening the performance of this kind of encapsulated detector.

Keywords: Compton-PET, lateral readout, SiPMs, monolithic crystals
Poster panel: 61

Poster Number:

Double Photon Emission Nuclides for Double Photon Coincidence Imaging  (#2778)

H. Takahashi1, A. Choghadi2, M. Uenomachi3, K. Shimazoe3

1 University of Tokyo, Institute of Engineering Innovation, Tokyo, Japan
2 University of Tokyo, Department of Bioengineering, Tokyo, Japan
3 University of Tokyo, Department of Nuclear Engineering and Management, Tokyo, Japan


We have proposed a new concept of time/position correlation type tomography method based on directionality sensitive gamma camera, which can identify incident gamma-ray direction. This tomography method utilizes the correlation between two gamma-ray photons and provides the radioactivity concentration in the body with high resolution, high sensitivity, and high signal to noise ratio. In-111 is known as a cascade gamma ray emission nuclide. Besides In-111, we have explored new candidates and found some other nuclides with different lifetimes.

Keywords: Double Photon Emission, Coincidence, gamma rays, Spectroscopy, Multi-nuclide Imaging
Poster panel: 64

Poster Number:

The feasibility Study of Novel flat panel detector using anthropomorphic phantom : Chest radiograph (#1095)

Y. Roh1, Y. Yoon2, K. Kim3, J. Kim3

1 Osong Medical Innovation Foundation, Integrated Medical Technology Team, Department of Research and Development, Medical Device , CheonJu, Republic of Korea
2 Kyushu University, Department of Health Sciences, Faculty of Medical Sciences , fukuoka, Japan
3 Korea University, Department of Health and Safety Convergence Science, Seoul, Republic of Korea


In the diagnostic radiology, there were many efforts to reduce the scatter radiation for better image quality. As a part of efforts, Authors have proposed the novel structure of indirect flat panel detector (FPD) system having net-like lead in the substrate layer, matching the ineffective area on the thin film transistor (TFT) layer to block the scatter radiation so that only primary X-rays could reach the effective area [Yoon et al.]. The feasibility of novel system was identified in previous study, therefore, we aimed to evaluate the performance of proposed novel FPD system using anthropomorphic phantom for the practical usage, especially the projection of plain chest radiography. To assess the performance of novel system, we calculated the number of incident photons (F2 Tally), the image contrast, and contrast to noise ratio (CNR) at two systems; the parallel grid system and the novel system. MCNPX 2.7.0 (Los Alamos National Laboratory, Los Alamos, NM, USA) software was used for the simulation and the setup of simulation was as below; 120 kVp with 2.5 mm Al filtration of beam quality, 180 cm of source to detector distance (SDD).
As the result of the number of incident photons to the detector, the ratio of novel system to the parallel system was 1 to 0.57. The image contrast of novel system was slightly lower than that of the parallel grid system (0.79 and 0.81). The CNR of novel system at soft tissue was 2.16 and that of parallel grid system was 2.50.
These results indicates that the novel system would be superior to the parallel system in perspective of the exposure dose for obtaining the equivalent image quality. The structure of novel system shows the possibility to reduce the exposure dose in plain chest radiography and this study would be also adapted to another radiographic examinations or modalities.

Keywords: MonteCarlo simulation, Chest radiograph, scattered radiation, radiation dose, image quality
Poster panel: 67

Poster Number:

Impact of Axial Ring Splitting on Image Quality for the Cost Reduction of Total-Body PET (#1331)

N. Efthimiou1, A. C. Whitehead2, M. Stockhoff3, C. Thyssen3, S. J. Archibald1, S. Vandenberghe3

1 University of Hull, School of Life Sciences, Faculty of Health Sciences, Hull, United Kingdom
2 University College London, Institute of Nuclear Medicine, London, United Kingdom
3 Ghent University, Department of Electronics and information systems, Ghent, Belgium


Total-Body PET (TB-PET) has high sensitivity, provides the means for simultaneous imaging of distant organs, offers fast kinetics using of short-lived isotopes and paediatric applications. Recently, the first TB-PET scanner was presented and the first results were impressive. However, the cost of a TB-PET scanner is prohibiting. Therefore we propose a flexible detector configuration which will increase the axial field of view and keep the overall cost low. We propose axial ring splitting, which will separate the full rings into rings with only even or only odd numbered detectors. In performance investigation, three configurations were considered (a) full rings (b) split rings (c) full rings - double in number. It was shown that config. (c) demonstrated the highest prompt count rate, as expected. Config. (a) was marginally better than config. (b). The same trends were observed in terms of noise equivalent count rate. Images reconstructed with OSEM (12 susets, 72 it.) demonstrated the improved image quality and noise properties of the longer scanners over classic PET.

Keywords: Total-Body PET, Monte Carlo, Image reconstruction, NECR, STIR
Poster panel: 70

Poster Number:

Development of a GPU-Based Mont Carlo Simulation Tool for PET (#1566)

Y. Lai1, Y. Zhong2, A. Chalise1, S. Zhou1, Y. Shao2, M. Jin1, X. Jia2, Y. Chi1

1 University of Texas at Arlington, Department of Physics, Arlington, Texas, United States of America
2 University of Texas Southwestern Medical Center, Department of Radiation Oncology, Dallas, Texas, United States of America


Monte Carlo simulation for positron emission tomography (PET) is important in terms of designing new prototype device and reconstruction algorithm. However, most of current simulation packages suffer from long simulation time. To fully address the time issue, a GPU based efficient and accurate simulation package, gPET, was developed and validated. gPET is built on the NVidia CUDA platform. The simulation process was modularized into three functional parts: 1) source management, including positron decay, transport and annihilation, 2) gamma transport inside the voxelized phantom, and 3) signal detection and processing inside the detector with repeatable part in three level hierarchy: panel, module and crystal scintillator. A predefined surface can be further used to simulate irregularly shaped detector. The performance of gPET was compared with GATE8.0 in two cases. 1) 1e7 positrons from C-11 point source centered in an 8cm3 cubic water phantom were simulated. Gammas generated from the annihilation process were further transported in an eight-panel-detector 6.65cm away from the center. 2) 1.5e7 gamma pairs were directly transported into a six-panel-detector with an inner cylinder shape. The mean positron ranges are 0.99mm and 1.14mm for gPET and GATE/Geant4, respectively. 0.5% difference in angular distribution is found for the gammas from annihilated positron. In both cases, the differences of energy distribution and spatial distribution over crystals are about 1.6% and 2.5%, respectively, in terms of the final coincidence pairs. The difference of the standard deviation of reconstructed image is below 0.2 mm. Corresponding execution time for gPET on single Titan Xp GPU (1.58GHz) and GATE8.0 on single Intel i7-6850K CPU (3.6GHz) were 20s and 4260s. In summary, gPET is an efficient and accurate Monte Carlo simulation tool for PET.

Keywords: PET, Monte Carlo simulation, GPU computing
Poster panel: 73

Poster Number:

Ultra fast prompt-gamma imaging for the online monitoring of the ion range in hadron therapy   (#1662)

J. Livingstone1, A. Etxebeste2, S. Curtoni1, D. Dauvergne1, M. Fontana3, M. - L. Gallin-Martel1, J. - M. Létang2, S. Marcatili1, C. Morel4, D. Sarrut2, E. Testa3

1 Laboratoire de Physique Subatomique et de Cosmologie, CNRS/IN2P3, Université Grenoble Alpes, Grenoble, France
2 CREATIS, Université de Lyon, Lyon, France
3 Institut de Physique Nucléaire de Lyon, CNRS/IN2P3, Université de Lyon, Villeurbanne, France
4 Aix-Marseille Univ, CNRS/IN2P3, CPPM, Marseille, France


Uncertainties in the ion range mean that the ideal ballistic properties of ions are not fully exploited in hadron therapy. Prompt γ imaging using a Compton camera has been proposed as a method of range verification for hadron therapy. Previous Monte Carlo studies of the CLaRyS Compton camera prototype demonstrated a 2 mm precision in the measured proton range for 108 particle histories using an iterative method of reconstruction of the γ emission profile. Using a hodoscope the reconstruction is simplified as for each detected event, the number of possible solutions is reduced to just two. Whilst the line cone reconstruction method is much faster than the iterative method, its precision was found to be worse due to the inability to discriminate between the two solutions. The aim of this study was to investigate the effect of temporal resolution on the precision of the range measured using the line cone reconstruction method and to propose a method of solution discrimination based on the time flight (TOF). To this end, GATE Monte Carlo simulations were performed. Protons of 160 MeV (109) were generated towards a cylindrical PMMA phantom. The GatePulseAdder was used to record coincidences between the scatterer and absorber stages of the CLaRyS camera. For each set of coincidences the line cone reconstruction was performed. For each solution, a TOF was estimated for the corresponding track (proton + γ) and compared to the TOF given by the simulation. The difference between the two, ΔTOF, was used to discriminate between the two solutions. The effect of the temporal resolution was investigated by placing an upper threshold on ΔTOF and the resulting emission profiles were compared to the real γ emission profile. It was found that a better temporal resolution resulted in a profile which more closely matches the real profile. It is envisaged to redesign the CLaRyS Compton camera with a fast scintillator such as CeBr3 to perform online range verification in hadron therapy.

Keywords: Compton camera, Monte Carlo simulation, GATE, Prompt gamma, Hadron therapy
Poster panel: 76

Poster Number:

Coupling X-ray and Light Transport for X-ray Luminescence Optical Tomography Simulations (#1836)

B. H. Zapien Campos1, A. Martínez Dávalos1, H. Alva Sánchez1, M. Rodríguez Villafuerte1

1 Universidad Nacional Autónoma de México, Instituto de Física, Mexico City, Mexico


In this work we study the feasibility of coupling X-ray and light transport in tissue using two methods: full Monte Carlo simulation, and a hybrid Monte Carlo for X-rays followed by a finite element solver for light transport. We show that both methods are equivalent through simulations in different phantoms, with the advantage that the hybrid method is ten times faster than full Monte Carlo.

Keywords: Biomedical imaging, optical tomography, Monte Carlo method., finite element method
Poster panel: 79

Poster Number:

Direct Annihilation Position Classification based on Deep Learning using Pair of Cherenkov Detectors: Monte Carlo Study (#1920)

K. Ote1, R. Ota1, F. Hashimoto1, T. Hasegawa2

1 Hamamatsu Photonics K.K., Central Research Laboratory, Hamamatsu City, Japan
2 Kitasato University, School of Allied Health Sciences, Sagamihara City, Japan


Cherenkov-based radiation detectors have been developed for time-of-flight (TOF) positron emission tomography. As Cherenkov photons are emitted in an extremely short time, their use can improve time resolution. However, because only up to 10 Cherenkov photons are yielded when a 511 KeV gamma ray interact with lead fluoride (PbF2) an accurate position estimation was difficult. Therefore, a 3D interaction position regression using a deep neural network (DNN) has been performed. However, a measurement of depth-of-interaction (DOI) as a label of supervised learning was difficult in practice. Thus, we propose a direct annihilation position classification using DNN on Cherenkov detector pair. Since the proposed method can directly estimate the annihilation position without using the DOI information of the detectors, we can easily perform supervised learning. The proposed method was evaluated by Monte Carlo simulation. For the simulations, a Cherenkov detector composed of monolithic PbF2 of 40 × 40 × 10 mm3 and a photodetector array were used. Two Cherenkov detectors were placed 50 mm face-to-face. The nine point-sources were placed 5 mm spacing between detectors along Z-axis. The annihilation position was classified using the DNN, whose inputs were detection positions on the photodetector plane (x, y) and timestamps of each photon. Training and validation datasets were generated while varying the single photon time resolution (SPTR) of photodetector. An accuracy of direct method was compared to an indirect method. The indirect method has two steps: (1) temporal and spatial interaction point is estimated for both Cherenkov detectors. (2) The annihilation position is calculated by TOF information along the line of response of the detector pair. In the SPTR σ = 10 ps, an accuracy ratio of the direct method to the indirect method was 0.93. Thus, the proposed method achieved a comparable accuracy to that of the indirect method and would ease the supervised learning.

Keywords: Gamma-ray detectors, Neural networks, Positron emission tomography, Supervised learning
Poster panel: 82

Poster Number:

The Napoli-Davis-Varna project for virtual clinical trials in X-ray breast imaging (#2216)

F. di Franco1, 2, P. Russo1, 2, G. Mettivier1, A. Sarno2, K. Bliznakova3, J. M. Boone4

1 Universita' di Napoli Federico II, Dipartimento di Fisica , Napoli, Italy
2 INFN, Sezione di Napoli, Napoli, Italy
3 Technical University of Varna, Varna, Bulgaria
4 University of California Davis, UC Davis Medical Center, Sacramento, California, United States of America


In order to overcome the limits of 2D mammography, 3D X-ray imaging techniques have been developed for reducing breast tissue superposition. Computed tomography dedicated to the breast (BCT) produces images of the uncompressed breast with isotropic spatial resolution and compared favorably to 2D digital mammography in differentiating soft tumor lesions. However, as an emerging technology, BCT needs a thorough dosimetric characterization as well as an appropriate optimization (spectra, geometry and components). In addition, its performance has to be compared to that of other imaging techniques, such as mammography and digital breast tomosynthesis (DBT). We present our project (Univ. Napoli, Davis, Varna) for virtual clinical trials based on a Monte Carlo (MC) platform for producing X-ray images (2D and 3D) of the breast anatomy and corresponding glandular dose estimates. We devised anthropomorphic digital breast phantoms derived from high-resolution clinical  breast CT scans acquired at UC Davis. 3D printed physical phantoms have been manufactured for experimental validation purposes, derived from clinical BCT scans.

Keywords: Virtual Clinical Trial, Breast Computed Tomography, Mammography, Monte Carlo simulation, Digital Breast Tomosynthesis
Poster panel: 85

Poster Number:

Monte Carlo Simulation Study of Annihilation Gamma Yield from C-11 and C-12 Irradiation in Water and PMMA Phantoms (#2388)

A. R. Chalise1, Y. Chi1, Y. Lai1, Y. Shao2, M. Jin1

1 University of Texas at Arlington, Department of Physics, Arlington, Texas, United States of America
2 UT Southwestern Medical Center, Department of Radiation Oncology, Dallas, Texas, United States of America


With technical advances in radioactive isotope generation, isotopes with positron decay (PDI) become a viable alternative to their stable counterparts for hadron therapy. Due to substantial signal boosting for positron emission tomography (PET), PDI is promising for accurate and precise online beam range (RB) verification. In this study, we quantified the online beam range verification ability of C-11 ion beams for PET imaging. Geant4 was used to simulate the dose and annihilation gamma (AG) distributions inside the phantom generated from C-11 and C-12 ion beams impinging water and PMMA phantoms. Quark Gluon String Pre-compound (QGSP), Binary Ion Cascade (BIC) and High Precision (HP) Neutron model reference physics list was used to govern the relevant physics processes for four therapeutic energies (ET ) (to reach the same penetration depth at each energy level for two carbon species). Post processing to obtain PET signal was performed with GATE and the retrace back strategy. AG yields with different acquisition times from start of ion history were compared between C-11 and C-12 beams. Exactly after the beam was turned on, AG with time tag <=5 minutes were collected and analyzed. For the water phantom at the lowest ET , the measured AG from C-11 was 8 folds higher than that from C-12 beam, while dropped to 1.09 folds for the highest ET . In general, with the increasing ET , AG yield rates increased for both species. Full width at half maximum (FWHM) of AG distribution peaks were 0.61 mm for C-11 and 6.01 mm for C-12 at the lowest ET . The results in PMMA were similar to those in water. The simulation results show that C-11 is an excellent candidate for accurate and precise online RB verification using PET.

Keywords: Hadron therapy, online beam range verification, C-11, PET, Monte Carlo Simulation
Poster panel: 88

Poster Number:

GEANT4 feasibility and effectiveness studies of a 3D-CZT single stage Compton Camera for BNCT therapeutic dose monitoring (#2497)

C. Gong1, S. Fatemi1, N. Auricchio2, E. Caroli2, S. Altieri1, 3, N. Protti1

1 National Institute of Nuclear Physics INFN, Unit of Pavia, Pavia, Italy
2 INAF-OAS Bologna, Area della Ricerca, Bologna, Italy
3 University of Pavia, Department of Physics, Pavia, Italy


Boron Neutron Capture Therapy (BNCT) is a binary radiotherapy based on 10B(n,α)7Li reaction, whose clinical outcome depends on the microscopic selective accumulation of 10B inside cancer cells. The 10B therapeutic dose can be monitored in vivothrough the measurement of the 0.478 MeV prompt gamma ray emitted by the de-excitation of 7Li.
The INFN 3CaTS project, funded by INFN (Italy), aims to realize this real-time dosimetry using an innovative 3D CZT drift strip detector. The traditional SPECT method relies on the use of high Z, highly attenuating collimators to get information about the flight direction and the point of origin of the detected photons thus heavily limiting the efficiency and spatial resolution of the imgaing system. However, the Compton Camera (CC) is an electronically collimated imaging system that uses the kinematics of Compton scattering, having higher spatial resolution and better performance than traditional SPECT method. This work focuses on the preliminary Monte Carlo feasibility simulations of a singlestage CC for BNCT therapeutic dose monitoring based on 3D CZT detectors. The relative efficienciesof the “true event” (defined as one Compton scattering followed by the photoelectric absorption of the scattered photon) in CZT with different volumes were simulated using Geant4. It shows that the relative efficiency of the “true event” in a 20 × 20 × 20 mmCZT detector reaches 17.7% at 0.478 MeV, a value comparable with other publications. At the same time, the CC imaging reconstructions are presently on-going to reconstruct the 10B dose distribution.
These preliminary results and possibilities for further studies will be shown in the talk.

Keywords: BNCT, Compton camera, Imaging reconstruction, Monte Carlo
Poster panel: 91

Poster Number:

Digital Zebrafish Phantom based on Micro-CT Data for Imaging Research (#2719)

M. Zvolsky1, N. Schreiner1, S. Seeger1, M. Schaar1, S. Rakers2, M. Rafecas1

1 Universität zu Lübeck, Institute of Medical Engineering, Lübeck, Schleswig-Holstein, Germany
2 Fraunhofer, Research Institution for Marine Biotechnology and Cell Technology, Lübeck, Schleswig-Holstein, Germany


Numerical simulations, including Monte Carlo techniques, are a powerful tool for characterising, evaluating, and optimising medical imaging devices and techniques. A vital aspect of simulations is a realistic phantom or model of the subject's anatomy. The zebrafish (Danio rerio)is becoming increasingly important in biomedical research, as it is a well-established model organism alternative to other vertebrate models such as mice and rats. The increasing interest of non-invasive imaging of the adult zebrafish necessitates the development of a realistic digital phantom. One of the goals of this study was to investigate how ex-vivo micro-CT scans can be used to create an emission phantom for nuclear imaging research. The present work describes the measurement and image processing strategies of the ex-vivo micro-CT scans. To find the best achievable soft tissue contrast, several fixation methods were compared, and the best CT scan parameters were identified. The fixation in formalin lead to the best soft-tissue contrast while preventing the fish from deforming during the measurement. The following regions of the zebrafish could be identified and segmented: Bones, eyes, brain, gills, heart, pharynx, spinal cord, liver, kidney, gonads, swim bladders, intestine as well as intestinal contents. However, some organs could only be distinguished by direct comparison with histological images, and some fatty and muscle tissue might have been mis-identified as parts of an organ. This is caused by the non-distinguishable attenuation coefficients of these organs at the employed X-ray energies. However, the established method for fish fixation and the optimised CT parameters lead to an excellent soft-tissue contrast comparable with images of high-magnetic-field MRI devices, thus enabling the segmentation of most organs of an adult zebrafish. The resulting preliminary digital phantom will be used for realistic nuclear imaging simulations. Further improvements are envisioned.

Keywords: Digital Phantom, Zebrafish, PET, Micro-CT
Poster panel: 94

Poster Number:

Evaluation of Near Infrared Fluoresence System for Dual Gamma-NIR Intraoperative Imaging. (#1228)

A. M. Almarhaby1, 2, J. Lees1, S. Bugby1, M. Alqahtani3, L. Jambi4, W. McKnight1, A. Perkins5

1 University of Leicester, Physics and Astronomy, Leicester, United Kingdom
2 Ministry of Health, King Fahd General Hospital, Jeddah, Saudi Arabia
3 King Khalid University, Radiological Sciences Department, Abha, Saudi Arabia
4 King Saud University, Radiological Sciences Department, Riyadh, Saudi Arabia
5 University of Nottingham, Radiological Sciences, Nottingham, United Kingdom


A novel hand-held hybrid optical-gamma camera (HGC) has previously been described that is capable of displaying co-aligned images from both modalities in a single imaging system. Here, a dedicated NIR imaging system for NIR fluorescence surgical guidance has been developed for combination with the HGC. This work has evaluated the performance of two NIR fluorescence imaging systems using phantom studies, various fluorophores and various experimental setups. The threshold detectable concentration of ICG and 800CW dyes were investigated for both systems. Bespoke micrometastatic lymph node phantoms and tissue-like layers were constructed to evaluate the detection capability. ICG could be detected at a minimum concentration of 1 µM for each camera. The lower thresholds for 800CW were 10-2 and 10-3 µM for the modified and NIR cameras, respectively. Both cameras were unable to detect micro-scale targets under 3 mm depth, but were able to identify larger targets as deep as 7 mm. Further improvements will combine one of these systems with the HGC for dual gamma-NIR fluorescence intraoperative imaging.

Keywords: Gamma camera, fluorescence imaging, ICG, IRDye® 800CW, cancer surgery
Poster panel: 97

Poster Number:

Development and Initial Performance Study of an All-digital TOF Brain PET System (#1670)

B. Zhang1, N. D’Ascenzo6, 4, C. - M. Kao5, L. Yang2, S. Liu2, L. Fang1, X. Zhang3, P. Xiao6, 4, Q. Xie6, 4

1 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
2 RaySolution Digital Medical Imaging Co., Ltd., Ezhou, China
3 The First Affiliated Hospital, Sun Yat-sen University, Department of Nuclear Medicine, Guangzhou, China
4 Wuhan National Laboratory for Optoelectronics, Wuhan, China
5 The University of Chicago, Department of Radiology, Chicago, United States of America
6 Huazhong University of Science and Technology, School of Life Science and Technology, Wuhan, China


Based on all-digital modularized PET detectors, which consist of 1:1 coupled LYSO/SiPM array and Multi-Voltage Threshold (MVT) digitizer, a Brain-PET scanner has been designed and developed quickly. The system contains 44 detector modules arranged on a 377 mm diameter ring, providing a transaxial field of view (FOV) of 320 mm and an axial FOV of 201.6 mm. A typical spatial resolution of 1.9 mm FWHM in center of FOV (with the method of 3D-OSEM-PSF) has been achieved. Aiming for higher performance and better image quality for neural-application, a sinogram-based TOF reconstruction method has been employed with the 551.7 ps FWHM time resolution. Initial performance of the system both for phantom and human have been evaluated and studied with and without time-of-flight (TOF) method, to characterize the benefit of TOF for enhance the contrast and reduce the noise.

Keywords: Brain PET, TOF, Performance
Poster panel: 100

Poster Number:

Performance evaluation of the RAYCAN digital small animal PET/CT system E180 (#1872)

N. D'Ascenzo1, Y. Ling1, Y. Hua4, L. Xiao1, A. Li1, E. Antonecchia1, 2, D. Xu4, Y. Liu1, R. Klein3, J. Knuuti3, J. Teuho3, Q. Xie1

1 Huazhong University of Science and Technology, Digital PET Lab, Wuhan, China
2 NEUROMED I.R.C.C.S., Medical Physics and Engineering, Pozzilli, Italy
3 Turku PET Centre, Turku University and Turku University Hospital, Turku, Finland
4 RAYCAN Technology Co., Ltd. (Suzhou), Suzhou, China


The design of a positron emission tomography scanner dedicated to functional imaging studies of small animals both in static and dynamic mode became nowadays essential to the study of novel drug models. We report a novel digital small animal PET system based on LySO/SiPM detectionmodules with digital large bandwidth readout. With respect to state of the art small animal PET/CT systems, E180 has a larger axial and transaxial field of view, which are 130 mm and 200 mm respectively, allowing total body small animal PET dynamic imaging for mice and rats. The system exhibits a NEC rate of approximately 1.5 Mcps, 1.45 Mcps and 0.48 Mcps for an activity of 81 MBq, 75 MBq and 60 MBq using the three NEMA-2008 phantoms respectively, a peak sensitivity of 7.5% and a space resolution of 1.3 mm at the center of the FOV. A unique feature of the digital micro PET/CT system E180 is the multi voltage threshold readout, which is facing the challenge of event rates and system electronics. Single events are collected by each detector module at a rate which is similar to other systems for the same injected activity. The challenge here is to obtain a digitization of the signal without introducing a significant fractional dead time. Furthermore, as the data stream includes all prompts, without any specific event selection, such as time and energy window, the additional dead time introduced by the conventional time coincidence electronics decreases significantly. As a result the digital PET/CT small animal scanner E180 will allow a pontentially innovative approach to the offline data analysis for the improvement of image quality, including, among others, optimization of the coincidence time window and energy window as a post-processing data analysis step, placing itself as a unique tool for the animal models functional imaging study.

Keywords: Small animal PET/CT, Digital electronics
Poster panel: 103

Poster Number:

Multi-Pinhole Collimator for ex-vivo Imaging of High-Energy Isotopes (#2070)

M. P. Nguyen1, M. C. Goorden1, F. J. Beekman1, 2

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


Imaging of radiolabeled molecules in biological tissue samples is important in biological and pharmaceutical research, e.g., to quantify drug distribution, to study metabolic pathways, to assess receptor distributions, and to localize enzymes or nucleic acids in cells. The recently developed EXIRAD-3D autoradiography technique (MILabs B.V.) with a dedicated focusing multi-pinhole collimator offers better than 120 µm (or 1.7 nL) spatial resolution images of 99mTc contained in cryo-cooled tissue samples. The available collimator for EXIRAD-3D is suitable for imaging SPECT isotopes with conventional-energy gamma emissions such as 125I (27 keV), 201Tl (71 keV), 99mTc (140 keV), and 111In (171 and 245 keV), while it is also of interest to image tissue samples containing isotopes with high-energy photon emissions. These include alpha and beta emitters such as 213Bi  and 131I that are important for theranostic applications, and positron emitters that result in 511 keV annihilation photons. This work aims to optimize and evaluate a new multi-pinhole collimator designed for this purpose through simulations. The collimator is first optimized by simulating a Derenzo phantom scan with a biologically realistic activity concentration of 18F at two system sensitivities (0.30% and 0.60%) for a series of collimators with different pinhole placement. The optimal collimators are subsequently evaluated for 18F scans with several activity concentrations and also for some other high-energy radioisotopes, namely 64Cu (511 keV), 124I (603 keV), and 131I (364 keV). Our results show that placing pinhole centers at a distance of 8 mm from the collimator inner surface yields optimal image quality. With the optimal high-energy collimators, EXIRAD-3D offers a resolution of 0.5 mm, 0.6 mm, 0.6 mm, and 0.45 mm when imaging 18F, 64Cu, 124I, and 131I, respectively, contained in tissue samples.

Keywords: multi-pinhole, SPECT, ex-vivo, high resolution, high energy
Poster panel: 106

Poster Number:

Trade-Offs Between TOF, Scan Time and Crystal Length on Image Contrast-to-Noise Ratio Performance for a Small Animal PET Scanner (#2256)

N. Zarif Yussefian1, M. Toussaint3, E. Gaudin2, R. Lecomte2, R. Fontaine1

1 Université de Sherbrooke, Interdisciplinary Institute for Technological Innovation 3IT, Sherbrooke, Québec, Canada
2 Université de Sherbrooke, Department of Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada
3 Université de Sherbrooke, Department of Computer Science, Sherbrooke, Québec, Canada


Recently, time of flight (TOF) scanners have become a mainstream in PET research owing to their ability to provide images with higher contrast-to-noise ratio (CNR). As the scintillation photon transport directly affects the timing resolution, decreasing crystal length has to be studied to improve timing performance even at the cost of sensitivity loss. This also improves radial spatial resolution and lessens the cost of design. Hence, we investigate the trade-off between TOF, scan time and crystal length with the goal of using TOF to compensate for CNR degradation caused by up to 19% decrease of scintillator volume. To do this, we used a model developed with GATE for the LabPET II scanner consisting of 1.12 x 1.12 x 10.6 mm3 LYSO scintillators. Simulations were performed following the NEMA NU4-2008 standards by varying crystal lengths, TOF resolution and scan time. The CASToR toolkit was used for reconstruction with MLEM for 0.3 x 0.3 x 1 mm3 voxel size. The CNR evaluation was performed with Matlab. Results showed that the same CNR value (4.2) is achievable for the scan time of 12 and 20 minutes if faster TOF resolution (100 ps rather than 200 ps) is incorporated for shorter scan. Furthermore, CNR was found to be equal if the same amount of statistics is incorporated by increasing the scan time for shorter crystal. Thus, TOF information can compensate for low statistics caused by shorter crystals. This may also be advantageous in achieving consistent CNR for lower fabrication cost related to the scintillator.

Keywords: CNR, TOF, Crystal Length, Scan Time, Trade-offs
Poster panel: 109

Poster Number:

Implementation of a Temperature Based Feedback System for Gain Stabilization in a MRI Compatible PET Insert  (#2363)

J. Afnan1, A. L. Goertzen2, 1

1 University of Manitoba, Canada, Biomedical Engineering Graduate Program, Winnipeg, Manitoba, Canada
2 University of Manitoba, Canada, Department of Radiology, Winnipeg, Manitoba, Canada


We previously reported on the hardware and software design of a slow control system for a MR compatible PET insert system. The PET insert is designed for simultaneous PET/MR imaging of small animals in a 7T animal MRI. The PET insert uses 16 detectors based on SensL SPMArray4B silicon photomultipliers (SiPMs) and dual-layer offset LYSO crystal blocks with crystal pitch of 1.27 mm. The slow control system is controlled by a BeagleBone Black microcomputer, which in turn connects to the host PC via Ethernet, allowing scaling to an arbitrary number of slow control units that each support 8 detector modules. In this work, we describe the implementation of a feedback control system to compensate for the SiPM gain variation caused by changes in temperature. We developed the feedback algorithm based on updating the detector bias voltage to ensure a fixed overvoltage, defined as the difference between bias voltage and detector breakdown voltage, as temperature changes. The feedback system was tested by placing the PET insert in a temperature chamber continuously cycling between 15°C and 25°C. When operated at a fixed bias voltage, the photopeak position varied by ±4.9%, while with dynamic bias voltage adjustment the variability was reduced to ±1.06%. Work is ongoing to characterize the stability of the PET insert in an MR environment.

Keywords: PET/MR small animal imaging, SiPM photomultipliers, temperature sensitivity of SiPM gain, preclinical PET/MR, SiPM gain compensation
Poster panel: 112

Poster Number:

Investigation of the Effects of Energy Window, Timing Window and Crystal Scatter on the Performance of the HiPET Preclinical PET Tomograph (#2429)

Z. Gu1, R. Taschereau1, D. Prout1, N. Vu2, A. Chatziioannou1

1 University of California, Los Angeles, Crump Institute for Molecular Imaging, Los Angeles, California, United States of America
2 Sofie Biosciences, Culver City, California, United States of America


HiPET is a recently developed high sensitivity and high resolution preclinical PET tomograph. The HiPET system employs a two layer LYSO/BGO phoswich depth of interaction (DOI) detector design, which also allows identification of the large majority of the cross layer crystal scatter (CLCS) events. In this work, the imaging protocol including energy window, coincidence timing window (CTW) and the CLCS event acceptance policy were investigated to achieve an overall optimized imaging quality.
The preliminary CTW used for the HiPET was 40 ns. An optimized CTW of 8/15/20 ns for LYSO-LYSO/LYSO-BGO/BGO-BGO coincidences was proposed based on the timing resolution of various event types. With the optimal CTW, the peak noise equivalent count rate (NECR) increased by 19% from 151 kcps to 179 kcps and 21% from 52 kcps to 63 kcps for the NEMA NU-4 mouse- and rat-sized phantoms, respectively.
The uniformity of the NEMA NU-4 image quality phantom improved significantly as the LLD_LYSO decreased, but was not significantly affected as the LLD_BGO decreased. The Recovery coefficients (RC) were significantly degraded when the LLD_LYSO was set to 150 and 200 keV, and were not significantly affected when the LLD_LYSO was equal or above 250 keV. The peak to valley ratio (PVR) across the 1.25 mm diameter rods in a reconstructed hot rod phantom image increased as the LLD_BGO increased, while a LLD_LYSO of 250 keV produced the highest PVR. An energy window of 250-650 keV for LYSO events and 350-650 keV for BGO events was proposed as the optimal energy window.
With the CLCS events included, the uniformity of the image quality phantom improved by 4% from 5.5% to 5.3%, and the peak NECR increased by 20% from 179 kcps to 215 kcps and 19% from 63 kcps to 75 kcps for the mouse- and rat-sized phantoms, respectively. While by excluding the CLCS events, the RC for the 1 mm diameter rod of the image quality phantom improved by 13% from 0.26 to 0.30.

Keywords: PET, energy window, timing window, crystal scatter, instrumentation
Poster panel: 115

Poster Number:

New detector designs for high resolution brain PET (#2591)

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

1 Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts, United States of America


The goal of this work is to design a detector with 2 mm intrinsic resolution and multiple levels of depth of interaction (DOI), suitable for brain dedicated Positron Emission Tomography (PET). Our starting point is a 2 cm thick LYSO:Ce detector with single-side readout, that contains laser induced optical barriers (LIOB) in a simple pixel like pattern extending half way through the crystal thickness. As this design comes with limitations related to complex DOI dependence and poor transversal resolution in the monolithic part, we are exploring two approaches to enhance performance beyond the basic design. These are: 1) Adding single photodetector elements to the sides and/or top of the detector, and 2) Keeping the single-side readout configuration but implementing a depth dependent pixel pattern where the pixel pitch varies throughout the crystal thickness. Our results show that the addition of side and top detectors can improve DOI performance by increasing the variety of metrics that can be used to extract DOI information. While side detectors may enhance transversal positioning, this approach comes with limitations in complexity of the event positioning. Our results further show that the increased number of channels by addition of side and/or top reflectors can be compensated for without compromising performance by relaxing the number of channels in the bottom array. The second studied approach is promising in that 2 mm transversal resolution is achievable throughout the detector thickness, except at the crystal edge close to the photodoetector plane, in combination with the possibility of extracting linear DOI information. A 53-60% difference between 4 and 16 mm interaction depth was observed by studying the maximum SiPM pixel signal relative to the rest of the array.

Keywords: PET, LIOB, optical transport, DOI, brain
Poster panel: 118

Poster Number:

Investigation of Highly-multiplexed SiPM Signal Readout for High-resolution TOF-DOI PET Detectors (#2731)

H. Park1, 2, J. S. Lee2, 3

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 Brightonix Imaging Inc., Seoul, Republic of Korea


 The aim of this study is to investigate the feasibility of a highly-multiplexed SiPM signal readout for developing high-resolution PET detectors with TOF and DOI capability.
The PET detector module is equipped with a two-layer staggered LSO crystal block and a 2×2 array of monolithic 16-channel TSV SiPMs (S13361-3050NE-04, HPK). The upper crystal layer consists of a 14×14 array of 1.78×1.78×8 mm3 crystals and the lower crystal layer consists of a 13×13 array of 1.78×1.78×12 mm3 crystals. Both the energy resolution and the 2D position map were obtained using four corner-node signals (i.e., A, B, C, and D) acquired from the 8×8 resistive charge division multiplexing circuit. The timing information was extracted using the multiplexed SiPM anode signals via a first-order high-pass filter. The data acquisition was performed using a DRS4-based high-speed waveform digitizer (DT5742B, CAEN). The detector performances were then comparatively evaluated with varying multiplexing ratio of 16:1, 32:1 and 64:1, respectively.
All the crystals with two-layer DOI information were clearly resolved in the 2D position map. Among all multiplexing schemes, the energy resolutions for each upper and lower crystal-layer were 10.5 ± 1.0% and 12.1 ± 1.7%, respectively. The 16:1 multiplexing yielded the best timing performance with CRT values of 325 ps (upper) and 342 ps (lower). However, the timing performances were kept almost constant even for the 64:1 multiplexing with CRT values of 336 ps (upper) and 347 ps (lower) thanks to the rapid baseline restoration of dark current of SiPMs based on the first-order high-pass filtering.
Based on the results, we concluded that the highly-multiplexed SiPM signal readout via the first-order high-pass filtering could be a promising candidate in developing high-resolution PET detector modules, greatly simplifying the data volumes to subsequent DAQ systems with moderate compromise in the TOF performance.

Keywords: High-resolution, Time-of-flight, Depth-of-interaction, Highly-multiplexed signal readout, PET detector
Poster panel: 121

Poster Number:

Characterization of Upgraded MADPET4 - A Small Animal PET Insert for a 7T MRI Scanner Using TOFPET2 ASIC (#2763)

N. Omidvari1, A. Zatcepin1, J. Cabello1, 2, S. Ziegler1, 3

1 Klinikum rechts der Isar, Technical University of Munich (TUM), Department of Nuclear Medicine, Munich, Bavaria, Germany
2 Siemens Medical Solutions USA, Inc. Molecular Imaging, Knoxville, Tennessee, United States of America
3 Klinikum der Universität München (LMU), Department of Nuclear Medicine, Munich, Bavaria, Germany


MADPET4 is a small animal PET insert for a 7 T MRI scanner, consisting of 2640 individually read out LYSO crystals, with 1.5×1.5 mm2 cross section area, arranged in two layers (6 mm and 14 mm). The insert was first developed and characterized with the PETsys readout system equipped with the TOFPET1 ASIC, using time-over-threshold (ToT) readout scheme. Small dynamic range of the TOFPET1 combined with high gain of the MADPET4 SiPMs and the ToT nonlinearity resulted in significant degradations in energy measurements. The readout electronics is now upgraded to use the TOFPET2 ASIC, offering a larger dynamic range and better linearity. In this work, TOFPET2 performance in charge integration mode is evaluated with high-gain SiPMs in MADPET4. The ASIC was tested in various modes, by changing its configurable parameters such as integrator gain, amplifier gain, integration window, and operating temperature. The image quality of the upgraded MADPET4 was evaluated using the NEMA NU4 image quality phantom and a micro-Derenzo phantom. Furthermore, since the insert has only a 2 cm axial FOV, resulting in a low sensitivity compared to other systems, possibility of using deep-learning based inter-crystal scatter (ICS) recovery methods were studied for sensitivity increase. The initial TOFPET2 characterization measurements with MADPET4 detectors showed linear response and improved energy measurement at different SiPM over-voltages, compared to TOFPET1. However, still some significant energy resolution degradations were observed in a number of channels, which were not resolved at any of the configured ASIC settings and require further investigation. The upgraded MADPET4 with the current configured ASIC parameters showed improved performance particularly in terms of spatial resolution, resolving the 1 mm rods in the Derenzo phantom. Further image quality enhancements are expected by including the ICS events recovered by the neural network, which is currently under optimization.

Keywords: PET, PET/MR, PET Insert, ASIC, Machine Learning
Poster panel: 124

Poster Number:

Feasibility study of photon counting detector for producing effective atomic number image (#1085)

N. Kimoto1, H. Hayashi1, T. Asakawa1, T. Asahara1, T. Maeda1, Y. Kanazawa2, A. Katsumata3, S. Yamamoto4, M. Okada4

1 Kanazawa University, Division of Health Sciences, Graduate School of Medical Sciences, Kanazawa, Japan
2 Tokushima University, Graduate School of Health Sciences, Tokushima, Japan
3 Asahi University, Department of Oral Radiology, Mizuho, Japan
4 Job Corporation, Yokohama, Japan


Currently, energy resolving photon counting detector (ERPCD) has been the focus of attention as promising technique in medical diagnosis. It can derive variable information related to an object such as effective atomic number (Zeff), etc. by analysis of X-ray energies. To accomplish highly accurate material identification, the response of a multi-pixel-type ERPCD should be taken into consideration because it distorts initial energy information. Additionally, we should note that there is a difference of beam hardening effect depending on Zeff. In this study, we propose a novel material identification method taking into consideration the detector response and beam hardening effect. Our method is based on analysis of attenuation factors of an object’s X-ray spectra in which the response function of a multi-pixel-type Cadmium Zinc Telluride detector is considered. The correction of the response function and beam hardening effect is applied to the attenuation factors. The corrected attenuation factors are converted into Zeff using our procedure in which the theoretical relationship between attenuation factors and Zeff is used as a reference. In order to evaluate robustness of our method, the following samples having mass thicknesses of 1 g/cm2 were measured using our proto-type ERPCD: acrylic, aluminum, bilayer structures of acrylic and aluminum with the ratios of 0.5:0.5, 0.67:0.33 and 0.8:0.2. The corresponding theoretical Zeffs are 6.5, 13.0, 10.5, 9.5 and 8.5. In addition, we measured real teeth. Zeff images for measured samples could be produced with an accuracy of Zeff +/- 0.5. Namely, our correction procedure can work properly, and the Zeff can be identified precisely. The Zeff of the teeth was distributed between Zeff = 9-15, and we were able to evaluate the fine structure such as enamel quantitatively. We believe that the quantitative information derived by Zeff image will provide an innovative tool for medical diagnosis in addition to conventional X-ray image.

Keywords: photon counting, response function, CZT, effective atomic number, material identification
Poster panel: 127

Poster Number:

Beam-filter-based dual-energy CT imaging by use of sinogram streaking (#1352)

S. Cho1, S. Lee1, J. Lee1, S. Cho1

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


CT scan by use of a beam filter placed between the x-ray source and the patient allows a single-scan low-dose dual-energy imaging with a minimal hardware modification to the existing CT systems.  We have earlier demonstrated the feasibility of such an imaging method with a multi-slit beam filter reciprocating along the direction perpendicular to the CT rotation axis in a cone-beam CT system. However, such method would face mechanical challenges when the beam filter is supposed to cooperate with a fast-rotating gantry in a diagnostic CT system. Increasing penumbra effects of the beam-filter edges would also add an additional burden. In this work, we propose a new scanning method and associated image reconstruction algorithm that can overcome these challenges. We propose to slide a beam filter that has multi-slit structure with the slits being at a slanted angle with the CT gantry rotation axis during a scan. A streak pattern would be created in the sinogram domain as a result. Using a notch filter in the Fourier domain of the sinogram, we removed the streaks and reconstructed an image by use of the filtered-backprojection algorithm. The remaining image artifacts were suppressed by applying l0 norm based smoothing. Using this image as a prior, we have reconstructed low- and high-energy CT images in the iterative reconstruction framework. An image-based material decomposition then followed. We conducted a simulation study to test its feasibility using the XCAT phantom, and showed a successful material decomposition.

Keywords: Low dose CT, Beam filter, Dual energy CT
Poster panel: 130

Poster Number:

Investigation of field modulation CT acquisition parameters for dose reduction and image improvement (#1575)

D. Kim1, H. Kim1, M. Lee1, H. - J. Kim1

1 Yonsei University, Department of Radiation Convergence Engineering, College of Health Science, Wounju, Republic of Korea

This work was supported in part by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2017R1A2B2001818).


Various dose reduction techniques have been studied in medical imaging. Region-of-interest (ROI) imaging is considered an effective method to reduce the exposure dose. We proposed ROI based field modulation acquisition for dose reduction and image improvement in computed tomography (CT) imaging system. A prototype of the CT system (TVX-IL1500H, GERI, Korea) was used. Field modulation CT images were reconstructed using a combination of truncated and non-truncated projection data. Four field modulation acquisition sets were investigated to derive appropriate acquisition parameters. Our proposed image restoration method based on a projection onto convex sets (POCS) algorithm corrected the truncated projection data. The image quality was evaluated using the contrast to noise ratio (CNR), and figure of merit (FOM) value. Among four acquisition sets, reconstructed image obtained by combining 672 truncated and 48 non-truncated projection data showed the best performance in terms of FOM values. In this study, we investigated proper acquisition parameter of field modulation CT imaging.

Keywords: Field modulation, ROI imaging, Computed tomography, Projection onto convex sets algorithm
Poster panel: 133

Poster Number:

The effect of dual-energy for microcalcification based on photon-counting spectral mammography (#1617)

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

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


The purpose of this study was to improve the discrimination of malignant and benign microcalcifications using dual-energy method. We used the photon-counting detector with energy discrimination capability, and the microcalcifications were classified using the appropriate dual-energy spectrum. In this study, CZT-based photon-counting spectral mammography system was modeled using GATE (Geant4 Application for Tomographic Emission) simulation tools. Two energy bins were used to obtain dual-energy images.
The ideal breast imaging should provide the high contrast, resolution to detect breast lesion due to the low attenuation characteristic for x-rays. Therefore, it is important to properly threshold the energy region for dual-energy acquisition. According to results, we confirmed that the noise is small and the contrast difference is large in the 26 to 30 keV for two types of microcalcifications. In addition, the classification performance of two types of microcalcification was improved when high contrast of 26-30 keV was included in low-energy region. Based on these results, we demonstrated that dual-energy subtraction can improve the contrast between malignant and benign lesions and can be well distinguished by automatic classification using score difference and segmentation methods. This study suggested the possibility of classifying microcalcifications based on spectral mammography to improve the diagnostic accuracy of breast lesions.

Keywords: Photon-counting spectral mammography, Microcalcification, Classification
Poster panel: 136

Poster Number:

Comparison of Ring Artifacts Removal by Using Neural Network in Different Domains (#2060)

W. Fang1, L. Li1

1 Tsinghua, Engineering Physics, Beijing, China


Ring artifacts in CT imaging has been blamed for many years. If not removed, these rings will severely deteriorate the image quality and affect doctor’s diagnosis results. Existed ring artifacts removal methods are generally based on filtering and averaging. In this paper, we propose a new ring artifacts removal method by using neural networks. We apply deep learning techniques respectively in image domain, projection domain and polar coordinate. Besides, in the past, ring artifacts removal is performed either in image domain or projection domain independently. In this paper, by incorporating reconstruction process into neural network, we firstly unite the image domain and projection domain processing for ring artifacts removal in the framework of deep learning. Traditional methods are implemented for comparison. Quantitative analysis is performed and it shows that the comprehensive neural network model including both image domain and projection domain information achieves best results.

Keywords: Ring artifacts removal, Residual deep learning, Comprehensive model, Photon counting detector, Spectral CT
Poster panel: 139

Poster Number:

Spectral CT Reconstruction via Low-Rank Representation and Texture Preserving Markov Random Field Regularization (#2312)

Y. Shi1, 2, Y. Gao2, X. Mou1, Z. Liang2

1 Xian'Jiaotong University, Electronic and Information Engineering, Xi'an, China
2 Stony Brook University, Radiology, stony brook, New York, United States of America


Photon-counting spectral computed tomography (CT) is capable of material characterization and can provide better diagnostic performance than traditional clinical CT. However, it still suffers from low photon counts for each individual energy channel which may cause serve artifacts in the reconstructed images. Since the images in different energy channels describe the same object, there are high correlations among different channels. To make full use of the inter-channel correlations and maintain the clinically meaningful texture information, this paper combined the tissue-specific texture prior with the low rank regularization to explore a superior texture preserving method for spectral CT reconstruction. Specifically, the high inter-channel correlations are modeled by low rank representation technique. The inner regional texture is characterized by a texture preserving Markov random field (MRF) prior. The proposed method integrates the spectral and spatial information into a unified Bayesian reconstruction framework. The Split-Bregman algorithm is employed to minimize the objective function to consider the non-differentiable property of low rank technique. Our approach was evaluated using a vivid phantom which can simulate the real tissue textures. As expected, the proposed method produced promising results in terms of not only preserving texture features but also suppressing image noise. Experimental results showed it outperformed the filtered back-projection (FBP) and low rank total variation (LRTV) prior methods in visual inspection and quantitative indexes such as RMSE, PSNR, structural similarity (SSIM) and feature similarity (FSIM).

Keywords: spectral CT reconstruction, low-rank, texture preservation
Poster panel: 142

Poster Number:

Potential of lower dose acquisition with Adaptive Statistical Iterative Reconstruction in computed tomography: A phantom study. (#2397)

I. Polycarpou1, D. Kaolis2, M. Andreou1, S. Papaspyrou1

1 European University Cyprus, Department of Health Sciences, Nicosia, Cyprus
2 Nicosia General Hospital, Department of Medical Physics, Nicosia, Cyprus


The increasing application of Computed Tomography (CT) in clinical practice leads to high cumulative doses of ionising radiation in patients. This issue has prompted a growing research interest in applications to reduce patient dose during CT imaging. Advanced reconstruction algorithms attempt to reduce patient dose whilst achieving comparable or even superior image quality compared to the standard CT protocol. ASIR is a commercially available reconstruction algorithm that results to lower noise and improved diagnostic confidence and therefore is a promising tool for acquiring data with lower dose. ASIR allows blending with the FBP algorithm at various levels. It is therefore important to determine the optimal level of blending with regards to the maximum dose reduction. Although, some studies have investigated the impact of ASIR in reduced dose protocols of CT images, the studies have so far focused on a limited range of blending levels of ASIR. This phantom study aims to comprehensively investigate the dependence of CT image quality with ASIR on the level of dose reduction. A thorax anthropomorphic phantom has been used and CT images were acquired for various dose and ASIR percentage levels (i.e. 0%, 10%, 20%, 30%, 40% and 50%). The performance of ASIR and dose reduction level have been quantitatively assessed in terms of noise modulation with standard deviation and noise power spectrum (NPS). The results of this study demonstrate that the combination of the ASIR algorithm and lower dose acquisition can reduce the radiation exposure of the patient without compromising image quality. A combination of 30 % DR and ASIR decreases the patient dose significantly while preserving the image quality compared to a standard protocol CT protocol.

Keywords: Adaptive Statistical Iterative Reconstruction, Computed Tomography, Dose Reduction, Image Quality
Poster panel: 145

Poster Number:

Development of Multi-channel Photon Counting X-ray Detector Using SiPM (#2673)

S. Hong1, J. Jung1, Y. Choi1

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


CZT and CdTe providing high count rate and energy resolution have been widely used as PCDs in x-ray imaging applications but they have serious disadvantages of high operating voltage, high cost and slow response. The purpose of this study was to develop a photon counting detector using SiPM, to design low noise and high-speed electronics and to assess the feasibility of the multi-channel detector based on SiPM for x-ray imaging (~150 keV). The detector was composed of a 1 x 8 array of 3 x 3 x 2 mm3 LYSO crystals coupled with a 1 x 8 array of 3 x 3 mm2 SiPMs. Multi-channel photon counting x-ray detector electronics was designed considering the features of fast and small x-ray signal output. Front-end board was designed using high speed, low distortion and low noise voltage amplifiers to improve SNR. 8 outputs were digitized using ADC having a low voltage differential signaling input/output for fast transmission and high SNR. In order to minimize not only distortion but count loss of the obtained x-ray energy spectrum in high flux x-ray application, pile-up rejection algorithm using arrival time difference, filtering to smooth noise signal and energy window to select a range of energy spectrum were implemented in FPGA. The USB 3.0 interface was used for data transfer to the host PC. To evaluate the performance of the detector and electronics in the energy region similar to x-ray, energy spectra were measured using Cs-137 (36 keV) and Co-57 (122 keV). In all channels, 36 and 122 keV peaks were clearly distinguished, and the energy resolution of 36 keV peak was 20%, and the energy resolution of the 120 keV peak was 25%. The measured results indicated that the proposed photon counting detector using SiPM had energy discrimination capability and feasibility for photon counting x-ray imaging. For further study, the development of 32 channel x-ray detectors using SiPM and performance enhancement such as energy resolution, count rate and noise of electronics are in progress.

Keywords: Photon counting, X-ray, SiPMs, Electronics
Poster panel: 148

Poster Number:

A prototype mobile C-arm CT system with high-resolution X-ray flat panel detector: performance evaluation and image optimization (#2736)

B. K. Cha1, S. Jeon1, C. - W. Seo2, K. Y. Ro3

1 KERI, Ansan, Republic of Korea
2 Yonsei University, WonJu, Republic of Korea
3 Genoray Co., Ltd., Seongnam, Republic of Korea


The advent of digital flat panel detector (FPD)-based C-arm CBCT imaging system provides an attractive technology for 2D image in standard fluoroscopy and 3D image with higher spatial resolution. A modern CBCT system with C-arm gantry incorporating X-ray flat-panel detector is widely used as a crucial imaging instrument for anatomical diagnosis and image-guidance in spine surgery, orthopedic and interventional suite. Higher resolution in dynamic radiological imaging such as fluoroscopy in C-arm CT imaging system is increasingly demanded by clinicians. Recently, the large-area CMOS flat panel imagers have been widely used in X-ray medical imaging applications including dental CBCT, mobile C-arm. The CMOS based X-ray detector has many benefits such as the higher readout speed, low noise, high spatial resolution and high system integration compared to amorphous silicon TFT-based flat panel detector. In this work, we have developed a prototype mobile C-arm CT system with flat-panel detector and successfully operated our system in an efficient way by controlling our C-arm gantry, X-ray pulse generator and CMOS-based flat-panel detector with high spatial resolution. The performance of CT imaging quality was carried out using the FDK reconstruction algorithm through acquired 2D projection images at different scanning angle and projection numbers. The quantitative analysis of image quality was investigated by using the cone-beam CT phantom (QRM GmbH) for contrast resolution, spatial resolution, noise and modulation transfer function (MTF). The 8lp/cm line pattern in spatial resolution was clearly visible. And low contrast resolution with 16mm diameter and 30HU(Hounsfield unit) were also detectable. The results indicated that the system was capable of providing better low contrast and spatial resolution. This paper will demonstrate the significant potential of X-ray CMOS flat panel imager in low-dose fluoroscopy applications for imaging-guided diagnosis and therapy.

Keywords: C-arm CT, and CMOS-based flat-panel detector, minimally invasive intervention, CT imaging quality
Poster panel: 151

Poster Number:

Rejection of coincidental noise events using multilayered semiconductor detectors (#1078)

Y. Kikuchi1, M. Horiuchi1, R. Okabe1, Y. Hiyama1, M. Shidahara1, S. Matsuyama1

1 Tohoku University, Graduate School of Engineering, Sendai, Aobaku, Japan


In this paper, a method for rejecting scatter and random coincidence (i.e. noise events) have been proposed to improve quantification of PET imaging. It is expected to be used for a novel type of PET scanner employing multilayered blocks of two-dimensional semiconductor detectors. The method is based on calculating the incidence angle of annihilation photons (Compton cone) using the principle of Compton camera imaging. Given that a detected event is true coincidence, the Compton cone is consistent with the LOR of the coincidental photon pair; on the other hand, a noise event results in the inconsistence. To evaluate the feasibility of the method, a simulation study was performed assuming a sixteen-head PET scanner system employing multilayered semiconductor detectors. GaAs and CdTe were chosen as the detector materials, and they were used to resemble the scattering detector and absorbing detector, respectively, in Compton camera imaging. Several configurations of the block with the different layer numbers were examined. As a result, approximately 60% of the scatter events were successfully rejected using blocks composed of GaAs (1mmt) × 40 layers and CdTe (1mmt) × 20 layers, and the percentage was changed according to the configuration. In addition, we set a margin for the rejecting of noise events with consideration for the limited accuracy of Compton cones, and the value of the margin significantly changed not only the percentage but also counts of true events that were faultily rejected together with noise events. With regard to rejecting of random events, every configuration in this study achieved success rates of approximately 90%.

Keywords: Semiconductor PET scanner, Three-dimensional position sensitive
Poster panel: 154

Poster Number:

Image Reconstruction With an Electron Tracking Compton Camera (#1133)

M. Inagaki1, K. Ogawa2, T. Tanimori3

1 Hosei University, Graduate School of Engineering, Tokyo, Japan
2 Hosei University, Faculty of Science and Engineering, Tokyo, Japan
3 Kyoto University, Faculty of Science, Kyoto, Japan


The electron tracking Compton camera (ETCC) system proposed by Tanimori, et al. (2006) facilitates reconstruction of three-dimensional distribution of radioisotopes without a mechanical collimator. This system has been developed for observations in astronomy, but it applicable to the imaging in nuclear medicine as well. The ETCC uses the information obtained by Compton scattered photons and recoiled electrons, which are generated in the process of Compton scattering. Without the information of the recoiled electron, the source position is localized only on a cone whose apex angle corresponds to the Compton scattering angle. However, as the ETCC can measure the energy and direction of the recoiled electron, the source position is fundamentally determined uniquely at a line on the cone surface. The quality of an ETCC image is affected by the uncertainty in the measurement of the energy of a scattered photon and that of the direction and energy of a recoiled electron. The former is affected by the Compton angle (Angular resolution: ARM) and the latter is affected by the dispersion on the cone with a Compton angle (Angular resolution: SPD). The paper studied the effect of the ARM and SPD on the reconstructed image. To evaluate the effect of the angular resolutions ARM and SPD, we assumed three detector geometries using one to three ETCCs. We assumed that the ETCC was composed of a CF4 gas detector with the size of 20×20×20 cm3 and NaI detector with the size of 20×20 cm2. The targeted gamma rays were 300 keV (In-111) and 4438 keV (Carbon therapy). The phantom used was water in sphere with a diameter of 3 cm. We transported photons with GEANT4 Monte Carlo code, and image reconstruction was performed with a list mode ML-EM algorithm. The results of simulations showed that the blurring of an image was improved with more than two detectors and the quality of images was better when we used high energy gamma rays.

Keywords: Compton camera, Monte Carlo simulation
Poster panel: 157

Poster Number:

A Feasibility Study of a Novel Variable-Aperture Full-Ring SPECT using Large-Area Pixelated CZT Modules: Simulation Results (#1290)

Y. Huh1, J. Yang1, O. Dim2, Y. Cui2, W. Tao3, Q. Huang3, Y. Seo1

1 University of California San Francisco, Physics Research Laboratory, Radiology and Biomedical Imaging, San Francisco, California, United States of America
2 Brookhaven National Laboratory, Department of Nonproliferation and National Security, Upton, New York, United States of America
3 Shanghai Jiao Tong University, Department of Nuclear Medicine, Ruijin Hospital, School of Biomedical Engineering, Shanghai, China


SPECT using a pixelated CZT detector (i.e., CZT-SPECT) has great potential to improve energy resolution, intrinsic spatial resolution, and sensitivity over those with a conventional NaI-SPECT system. We propose a CZT-SPECT concept for brain and general purposes designed with a variable-aperture full-ring geometry to overcome the limited field of view and to minimize the overlapping data between each detector using virtual projection data. We evaluated our SPECT system concept with analytical phantom images acquired from Monte Carlo (MC) simulations and iterative image reconstruction. The proposed SPECT system consists of eight modules of 153.6 mm × 153.6 mm large-area pixelated CZT detectors with energy-optimized parallel hole collimator. Each detector module can move radially in and out, and rotate around the center of the detector. Each virtual projection data for reconstruction can be generated by combining two projection data taken by two detector modules. Spheres of different sizes, modified sphere, and Derenzo-like hot spot phantoms were simulated to compare with a conventional SPECT system and the phantom images were reconstructed using the STIR package. We obtained the clear reconstructed images of a typical sphere phantom study from conventional and proposed systems. The reconstructed images of modified sphere and Derenzo phantom studies, whose rods are located outside of FOV, were not truncated due to the virtual projection data and small size rods could be identified even though there were some artifacts due to non-optimized conditions. Initial simulation studies demonstrate that our proposed SPECT system with expandable FOV using CZT detector modules is a reliable concept. For further study, the optimal geometry of the system and virtual projection data will be investigated to improve the sensitivity and spatial resolution and additional performance improvements will be developed such as reducing the number of projection data needed for reconstruction.

Keywords: CZT, SPECT, Monte Carlo, STIR
Poster panel: 160

Poster Number:

DOI vs. TOF: performance comparison of two helmet-type PET prototypes based on the NEMA NU2 standards (#1510)

G. Akamatsu1, E. Yoshida1, H. Tashima1, Y. Iwao1, H. Wakizaka1, T. Maeda1, M. Takahashi1, T. Yamashita2, T. Yamaya1

1 National Institute of Radiological Sciences (NIRS-QST), Chiba, Japan
2 ATOX Co., Ltd., Tokyo, Japan


Brain PET imaging plays key roles in neurology, neuro-oncology, and molecular imaging research fields. To fulfill a growing demand for a high-sensitivity, high-resolution, low-cost and compact brain PET system, we have proposed a helmet-shape detector arrangement. In the 2016 conference, we described development of the first prototype, which had depth-of-interaction (DOI) detectors without time-of-flight (TOF) capability. Last year, we reported the second prototype which had TOF capability of a 245-ps coincidence timing resolution without DOI capability. The crystal thickness was optimized through simulation so as to reduce the parallax error while keeping the gain in image quality. The detector was composed of 12×12 lutetium fine silicate (LFS) scintillation crystals (4×4×10 mm3) connected to a 12×12 multi-pixel photon counter (MPPC) array with one-to-one coupling.  In this work, we have evaluated performance of the second prototype based on the NEMA NU2 standards and compared the results with those of the first prototype. The values measured with the first (DOI) and second (TOF) prototypes were: spatial resolutions of 2.8 and 4.3 mm at the central position (1-cm offset) by the filtered back-projection method; sensitivities of 13.4 and 4.1 kcps/MBq; scatter fractions at the low activity of 31% and 31%; and peak noise-equivalent count rates (NECR) of 20.0 kcps at 2.6 kBq/mL and 18.6 kcps at 9.0 kBq/mL. With a TOF sensitivity gain of 5.4 (= 20 cm diameter / 3.675 cm TOF localization), effective sensitivity and effective peak NECR were 22.3 kcps/MBq and 101.2 kcps at 9.0 kBq/mL for the second prototype. In the image quality assessment, the second prototype outperformed the first prototype. Although the first prototype had better performance for spatial resolution and sensitivity themselves which were measured with low count rate, the TOF capability of the second prototype worked well to reduce the random ratio, which resulted in the superior image quality.

Keywords: PET, Brain, DOI, TOF
Poster panel: 163

Poster Number:

Performance evaluation of the TRIMAGE brain PET system (#1774)

G. Sportelli1, 2, M. G. Bisogni1, 2, N. Camarlinghi1, 2, P. Carra1, 2, A. Del Guerra1, 2, M. Morrocchi1, 2, A. Pilleri1, 2, N. Belcari1, 2

1 University of Pisa, Department of Physics , Pisa, Italy
2 Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, Pisa, Italy


We present the characterization results obtained with the TRIMAGE PET detectors and the related electronics. The system is designed to feature better PET performance than conventional clinical PET/MR. It consists of a ring of 18 sectors in the form of rectangular detectors, 55 mm (transversal) × 163 mm (axial) size, big enough to image a human brain. Early characterization indicates a sensitivity of 6.8% at the centre of the field of view with an energy window of 350 - 650 keV and a CTR of 690 ps. The overall imaging performance evaluation is based on the National Electrical Manufacturers Association (NEMA) standards NU2-2018 or NU4-2008, whenever appropriate.

Keywords: Brain PET, molecular imaging, PET/MR, PET/MR/EEG, TRIMAGE project
Poster panel: 166

Poster Number:

Performance evaluation of the adjustable gantry PET for small animal PET imaging (#1953)

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

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


Adjustable gantry PET(AGPET) was developed for small animal PET imaging. AGPET consisted of four detector heads arranged in the square gantry. The field of view (FOV) can be adjustable depending on the size of the object by transforming the location of detector heads. A single detector module was composed of 12 x 24 arrays of LYSO crystals which have 2.1 x 2.1 x 20 mm3 of dimension. The detector module was optically coupled to two 8 x 8 SiPM arrays. The detector module was connected to a signal processing board which multiplexed the signals into 128:4 of positioning information. Two detector modules form one detector head. At each detector module, a depth dependent reflector pattern was applied to obtain depth of interaction(DOI) information. To investigate performances of the AGPET, two detector blocks were implemented. Intrinsic spatial resolution was measured with a Na-22 by moving 0.3 mm of steps at median line of the detector block. Point source was imaged by varying gantry size for 30, 50 and 90 mm of FOV. Full width half maximum (FWHM) was evaluated with and without DOI information. The intrinsic spatial resolution ranged from 1.2 mm to 1.95 mm with an average of 1.7 mm. The point source image according to the FOV size and DOI information will be shown in the presentation.

Keywords: Small animal PET, Depth of interaction
Poster panel: 169

Poster Number:

On-Beam Imaging of 718 keV Prompt Gamma using Si/CdTe Compton Camera for Carbon Ion Beam (#2037)

R. K. Parajuli1, M. Sakai1, W. Kada2, M. Kikuchi1, K. Arakawa1

1 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. Although carbon ion radiotherapy can deliver high-precision cancer treatment, there are still some uncertainties due to technical, physical, and biological optimizations that require further developments. During beam irradiation on the target, the primary beam stops inside the target, and the nuclear reaction yields excited nuclides and generates high-energy gamma radiations. Since the gamma ray emissions and the Bragg peak are correlated, understanding precisely the beam activity by monitoring the gamma ray emissions is essential. In this study, we perform imaging of 718 keV prompt gammas emissions for the Carbon ion beams (290 MeV/u) used for clinical treatment purpose. A Silicon/CadmiumTelluride detector based Compton camera was used for beam monitoring and a simple backprojection-imaging algorithm was used to develop Compton images. The energy spectrum obtained through the Compton camera shows both the energy peaks of 511 keV and 718 keV gamma ray events. In our previous studies, we have discussed about the 511 keV annihilation gamma imaging using our Compton camera. In comparison with the energy spectrum of 511 keV annihilation gammas, intensity of 718 keV prompt gammas was relatively low. However, 718 keV prompt gammas are not prone to washout effect and therefore could provide valuable information regarding on-beam monitoring. We were able to develop a preliminary Compton image using 718 keV gamma events and were suitable for further statistical analysis. The 718 keV gamma emissions were due to the de-excitation in 10B* resulting from the product of β+ emissions of 10C. Despite the low production of 10B*, if the 718 keV gamma emission could be measured effectively, it would be an innovative breakthrough in hadron therapy.

Keywords: Beam monitoring, Carbon-ion radiotherapy, Compton Camera, Prompt gamma, Annihilation gamma
Poster panel: 172

Poster Number:

Simulation study comparing open geometry cardiac TOF-PET system with full ring scanner (#2088)

S. Oliver1, L. Moliner1, J. Prats1, M. J. Rodríguez-Álvarez1

1 Universitat Politècnica de València-Consejo Superior de Investigaciones Científicas, Instituto de instrumentación para imagen Molecular, València, Spain


The aim of this work is to compare simulation results of a phantom located at estimated heart position in a cardiac TOF-PET system, with the same phantom in a full ring PET scanner. With this study, we pretend to determine the TOF resolution needed to achieve reasonable image quality for the cardiac TOF-PET system. The proposed geometry for cardiac TOF-PET system is an open geometry that consists of 36 modules arranged in two perpendicular sets of two parallel plates each one. The modules are distributed in a matrix in each plate.  We performed Monte Carlo simulations for the proposed cardiac PET system with different TOF resolutions, in order to test image quality improvement. Moreover, we performed simulations for a full ring scanner with the same TOF resolution values to compare both results. Our results show that increasing TOF resolution reduce distortion and artifact effects and that TOF resolution of the order of 200 ps is needed to reduce artifacts generated by the open geometry of the simulated cardiac PET system.

Keywords: Positron emission Tomography, cardiac PET system, imaging reconstruction
Poster panel: 175

Poster Number:

Development of the uEXPLORER singles acquisition framework (#2267)

A. R. Selfridge1, J. Li3, R. Tao3, S. Chu3, W. Liu3, E. Leung2, B. Spencer2, R. D. Badawi2, S. R. Cherry1, 2

1 University of California, Davis, Biomedical Engineering, Davis, California, United States of America
2 University of California, Davis, Radiology, Davis, California, United States of America
3 United Imaging Healthcare, Molecular Imaging, Shanghai, China


Commercial whole-body PET/CT scanners only output the processed coincidence list mode data or sinograms necessary for image formation. While this approach simplifies the clinical workflow and meets the diagnostic requirements for doctors and patients, it places limitations on researchers and data scientists looking to develop new techniques for PET data processing. Electronics and data acquisition systems for reading out single events, in contrast, have been developed for prototype systems, but have not been translated into clinical systems. The system electronics of the uEXPLORER total-body PET system were designed to provide two data acquisition paths to simultaneously collect raw single event data and processed coincidence data. In collaboration with United Imaging Healthcare we have developed a software platform to acquire singles data simultaneously with the coincidence data natively produced by the system. We have also developed and validated a workflow for processing single event data, which includes merging data streams from individual PET rings, coincidence sorting, data correction, and reconstruction. Measurements using a uniform 68Ge phantom show 0.6% more coincidences for data acquired with the singles based data path, compared to the online coincidence path. This minor difference reflects behavior of the system firmware that is not easily implemented in the offline coincidence processing software. Sinograms did not otherwise show artifacts indicating differences in the behavior of the two data paths. Measurements using a NEMA IQ phantom demonstrate that the singles data workflow yields quantitatively accurate reconstructed images and coincidence data that is compatible with the vendor’s workflow. This secondary data path enables a range of studies evaluating the possible role of singles data in data correction and alternative coincidence sorting methods.

Keywords: uEXPLORER, total body PET, singles acquisition
Poster panel: 178

Poster Number:

Imaging performance evaluation of a multiple-isotope PET with 44mSc tracer (#2488)

T. Fukuchi1, M. Shigeta1, H. Haba2, S. Yamamoto3, Y. Watanabe1

1 RIKEN, Center for Biosystems Dynamics Research, Kobe, Japan
2 RIKEN, Nishina Center for Accelerator-Based Science, Wako, Japan
3 Nagoya University, Graduate School of Medicine, Nagoya, Japan


We have been developing a new PET system for multiple-tracer imaging, called a multiple-isotope PET (MI-PET). MI-PET can identify the tracer by detection of the prompt γ-rays emitted after positron emission. MI-PET is composed of a PET system and additional γ-ray detectors. The PET system consists of pixelized GSO scintillation detectors and has a ring geometry with 95 mm in inner diameter. Eight additional detectors are mounted on each side of the PET ring. For the multiple-isotope imaging using MI-PET, at least one positron-γ emitter is necessary as a tracer. Scandium-44 is one of promising candidate of specific tracer for MI-PET, because of its large positron and prompt γ-ray emission ratios and moderate prompt γ-ray energy (1157 keV) and half-life (44Sc: 4.0 hours, 44mSc: 58.6 hours). In order to evaluate imaging performance of MI-PET with 44mSc as a second tracer, we performed a dual-isotope phantom and animal experiments. For the phantom experiment, we prepare 18F (pure positron emitter) and 44mSc dissolved in water, and both and each isotope were poured into three rods. A 30-min scan was performed simultaneously for three rods. From the reconstructed images with the absence or presence of the prompt γ-ray detection, we can clearly observe 18F and 44mSc distributions separately. The spatial resolution of the second tracer (44mSc) is comparable to the intrinsic MI-PET resolution. In order to evaluate the practical performance of MI-PET with 44mSc tracer, we conducted a dual-isotope mouse experiment. In this experiment, 18F-FDG and 44mSc were administered to an ICR male mouse by tail vein injection. A 30-min scan with bed motion was performed under anesthesia. From the reconstructed images, we can clearly see difference of 18F-FDG and 44mSc distributions. Therefore, we successfully demonstrated the feasibility of the practical dual-isotope imaging. In future, we will synthesize useful MI-PET drug labeled by 44mSc and perform multiple-drug analysis.

Keywords: positron emission tomography, multiple-isotope imaging, gamma-ray
Poster panel: 181

Poster Number:

PET system cooling with vortex tubes (#2628)

A. R. Selfridge1, J. Bec1, R. D. Badawi2

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


Cooling and heat dissipation are key aspects of PET detector design which are often addressed by the use of mechanically refrigerated air chillers. Vortex tubes, however, are an alternative source of chilled air which are well suited to PET applications based on their compact size, fast response, and large cooling capacity.
We have evaluated the performance of a cooling system based on vortex tubes to determine their viability for PET applications. Chilled air was directed towards a detector stack which we have designed for use in preclinical PET/MRI, where temperature was measured at the back face of an SiPM array. Performance was assessed based on temperature stability, transient response, and detector crystal separation.
Stability of the cooling system was determined by recording temperature for an extended period while the detector was powered. Standard deviation of the temperature was less than 0.06℃ over four hours, despite the high density of frontend electronics near the SiPM arrays. Transient response to a step change in the temperature set point was similar for both the vortex tube and mechanical chiller, with response times of 218 and 225 seconds. Finally, crystals were well resolved across a range of temperature set points for the vortex tube, reflecting both the stability of the detector and the practical impact of the cooling system on detector performance.
Geometric constraints of PET/MRI inserts make incorporation of mechanical chillers difficult. With performance similar to or exceeding mechanical air chillers, vortex tubes are well suited these and other PET applications where their flexibility and compactness can significantly simplify system design.

Keywords: PET system cooling, vortex tube, PET/MRI
Poster panel: 184

Poster Number:

A novel technology of multi-pinhole SPECT for Human Myocardial Perfusion Imaging (#2706)

C. Zhang1, H. Wang1, D. Wang1, T. Xu1, N. Jiang1, Y. Liu2

1 Beijing Novel Medical Equipment Ltd Beijing China, Medical Imaging, Beijing, China
2 Tsinghua University, Department of Engineering Physics , Beijing, China

perform the experiment and write this summary


Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) remains critically important for diagnosing and evaluating treatment of patients with coronary artery disease (CAD). However, parallel-hole collimator suffers from prolonged imaging times because of relatively low detection efficiency. In this study, we acquired projection data of patient MPI for rest and exercise with 12-pinhole collimator and Siemens T series. The myocardium is not sampled sufficiently using only one view of a multi-pinhole (MPH) collimator. Therefore, we scanned two cases:(a) single view from -170of main turning device, and (b) 2 views from -170and -135. Patient MPI was imaged with Siemens T and SPECT ImageE NET 632 to assess the image quality of both systems. In comparison with the parallel-hole collimator, the performance of a single view of the 12-pinhole collimator exhibits a substantial loss in spatial resolution in anterior wall myocardial. Also the myocardial image is distorted especially at the apex with the 12-pinhole collimator. The reconstruction from 2 views presents considerable improvement in terms of the image quality compared with the reconstructions from 1 view. The apex does not show the pointed shape observed with 1 view. Slight non-uniformities are visible in the apex, but the images overall are very well resolved. The work presented here shows that performing MPH imaging with 2 views can significantly improve the quality of the reconstructed image from that obtained with just a single view. Therefore, we conclude that MPH MPI offers significant advantages over rotational SPECT with conventional parallel-hole collimators, and further it is worth investigating MPH collimator for tomographic MPI.

Keywords: SPECT, multi-pinhole collimator, myocardial perfusion imaging, angular sampling
Poster panel: 187

Poster Number:

OMEGA - Open-source MATLAB Emission Tomography Software (#1106)

V. - V. Wettenhovi1, M. Vauhkonen1, V. Kolehmainen1

1 University of Eastern Finland, Department of Applied Physics, Kuopio, Finland


Open-source MATLAB Emission Tomography Software (OMEGA) is a toolkit for MATLAB, designed for efficient and easy image reconstruction of Positron Emission Tomography (PET) data. The goal of OMEGA was to have an easy to use software for MATLAB that would also be computationally efficient to use. OMEGA includes 10 maximum likelihood algorithms, 7 maximum a posteriori algorithms and 9 priors. Any number of these algorithms and/or priors can be run at the same time. The reconstructions can be performed either by using purely MATLAB commands, using C++ MEX-files or OpenCL C/C++ MEX-files, where the OpenCL method allows the use of any GPU or CPU. The OpenCL method is furthermore completely matrix free, allowing the reconstructions to be performed even with devices having only a few gigabytes of memory. The toolkit additionally includes direct support for GATE output data in ASCII, LMF or ROOT format, allowing easy reconstruction of simulated GATE data. Additional features include support for dynamic imaging, support for both sinogram formatted data as well as raw list-mode data, operators for forward and backprojections and automatic randoms correction from delayed coincidences. Furthermore, several GATE specific features are present like the ability to extract and reconstruct only true coincidences.

Keywords: image reconstruction, PET, GATE, MATLAB, GPU
Poster panel: 190

Poster Number:

RISE: Tomographic Image Reconstruction in Positron Emission Tomography (#1221)

C. Lemesios1, L. Koutsantonis1, C. Papanicolas1, 2

1 The Cyprus Institute, Computation based Science and Technology Research Center, Aglantzia, Cyprus
2 National and Kapodistrian University of Athens, Physics Department, Athens, Greece


We present results from the application of RISE in Positron Emission Tomography (PET). The RISE implementation in PET is evaluated in this report using a 3-D Shepp-Logan software phantom featuring seven 3-D ellipsoids. The dataset was produced on a cylindrical PET scanner, stored in list-mode-format. Using the appropriate image quality metrics (CC, SSIM), we compare our results to those obtained with the Maximum Likelihood Expectation Maximization (MLEM) method. The RISE results compare favorably to those of MLEM.

Keywords: PET, reconstruction, RISE, AMIAS
Poster panel: 193

Poster Number:

Anatomy Guided Reconstruction Using l1  Bowsher Prior (#1307)

S. K. Kang1, J. S. Lee1

1 Seoul National University, Department of Biomedical Sciences, Seoul, Republic of Korea


In comparison with CT and MRI, PET has relatively poor spatial resolution and noise characteristics. Incorporating detail anatomical information provided by CT or MRI into PET image reconstruction is an actively investigated method to overcome these shortcomings of PET.  The original Bowsher’s method that is based on second order smoothing prior sometimes suffers from over-smoothing of detailed structures. Therefore, in this study, we propose l1 norm based Bowsher prior and closed solution for the iterative image reconstruction based on this new prior.
We first implemented penalized reconstruction with original (l2) Bowsher prior using separable surrogate function. Then, modified proximal gradient algorithm was devised because the reconstruction scheme from the original Bowsher prior was not applicable to the proposed prior, which was convex but not smooth function. Both reconstruction method were accelerated by ordered subset algorithm. Three-dimensional brain phantom was generated for 18[F]-FDG model. The brain phantom consisted of four different regions, gray matter (GM), white matter (WM), large tumor and small tumor. Bias and standard deviation were calculated for these regions.
The proposed l1 Bowsher prior recovered well the detailed structure of GM and various tumors under the high-level noise circumstances. Although the original Bowsher prior over-smoothed small tumor, proposed prior preserved the shape and intensity of the small lesion. The quantification results show that the proposed prior and reconstruction yields the best performances in terms of accuracy for all regions.
In conclusion, we proposed novel anatomy guided prior yielding better quantification of tumors as well as the GM and WM than the previous approaches. Modified proximal gradient algorithm was devised to solve the proposed non-smooth prior.

Keywords: PET/MRI, Penalized reconstruction, Bowsher prior
Poster panel: 196

Poster Number:

DualRes-UNet: Limited angle Artifact Reduction for Computed Tomography (#1378)

T. Zhang1, 2, H. Gao1, 2, Y. Xing1, 2, 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 (Tsinghua University), Beijing, China


In some special cases of computed tomography (CT), limited angle problem occurs due to the limitation of imaging configuration and system design, which could cause severe artifacts in reconstructed CT images. In this work, we explore a Convolutional Neural Network (CNN) architecture for limited angle artifact reduction in CT imaging, named as DualRes-UNet. In our network, we adopt continuous down-sampling layers similar to U-Net to obtain a large receptive field view so that it can capture high level structure of object. Like concatenation operations, the proposed DualRes modules introduce high resolution features into the up-sampling process, which are beneficial to preserve more details and textures in final images from CNN. Preliminary experiments of 120~150 degree of limited-angle CT were conducted, which demonstrate that our DualRes-UNet can effectively suppress the limited angle artifact and improve the quantitative accuracy of CT images.

Keywords: Limited Angle, Computed Tomography, Artifact Reduction, Convolutional Neural Network
Poster panel: 199

Poster Number:

Machine-Learning-Based Filtering of List-Mode Data in Positron Emission Tomography (#1463)

J. Grahe1, 2, A. Vieth2, A. Salomon1, D. Schug2, T. Dey2, A. Goedicke1, V. Schulz2

1 Philips Research, Oncology Solutions, Eindhoven, Netherlands
2 University of Aachen, Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, Aachen, Germany


We propose to filter list-mode data acquired in positron emission tomography (PET) using machine-learning classifiers to increase the image quality by reducing the amount of scattered and random coincidences.
Filters based on supervised learning are trained using a Monte-Carlo simulation to classify true, scattered and random coincidences based on the energies of the gamma particles, the time-of-flight (TOF) information and the accumulated attenuation along the line-of-response (LOR).
The simulation software GATE is used to accurately model these parameters based on the used activity and attenuation distribution in the NEMA IQ phantom as well as the properties of the employed Philips Vereos scanner.
Classifiers are trained and validated using the generated Monte-Carlo ground truth to predict the true coincidences in a separate test data set containing 55% true, 25% scattered and 20% random coincidences.
Only coincidences classified as true are used in the reconstruction.
The multi-layer perceptron (MLP) and the gradient boosting classifier (GBC) filter 86.1%/86.3% of the random, 50.7%/52.3% of the scattered and 6.1%/7.2% of the true coincidences, respectively.
Both classifiers perform equally well in regard to their receiver operating characteristics.
The quantitative evaluation of the images reconstructed from filtered data requires a correction for the remaining random and scattered coincidences passing the filter, which is not implemented yet.
However, images reconstructed using the filtered coincidences and images reconstructed using all coincidences, but with scatter and random corrections, exhibit similar characteristics when compared visually.
Backprojections of the filtered coincidences confirm the good classification of scattered and random coincidences by the filters.

Keywords: image reconstruction, machine-learning, TOF, PET, image quality
Poster panel: 201

Poster Number:

Sparse-View Tomography via Displacement Vector Field Interpolation (#1442)

G. L. Zeng1, 2

1 Weber State University, College of Engineering, Applied Science and Technology, Ogden, Utah, United States of America
2 University of Utah, Department of Radiology and Imaging Science, Salt Lake City, Utah, United States of America


Sparse-view tomography has many applications such as in low-dose computed tomography (CT). Using under-sampled data, we are not expecting to obtain a perfect image. Our goal is to obtain a tomographic image that is better than the naïve filtered backprojection (FBP) reconstruction that uses linear interpolation to complete the measurements. This paper proposes a method to estimate the un-measured projections by using non-rigid deformation. This is a non-linear method and the interpolation is performed on the displacement function (instead of the sinogram itself). The proposed method is compared with the interpolation methods. The proposed method shows superior performance.

Keywords: Limited data imaging, tomography, estimation
Poster panel: 202

Poster Number:

Study of the impact of patient movement on penalized maximum-likelihood image reconstruction for a limited-angle PET scanner (#1545)

H. Zhang1, M. Saad1, S. Abbaszadeh1

1 University of Illinois at Urbana Champaign, Urbana, Illinois, United States of America


Patient movement can cause artifacts and image quality degradation in positron emission tomography (PET) imaging. We are developing a two-panel high-spatial-resolution PET system dedicated for head and neck cancer imaging. To mitigate the problem of limited-angle artifacts, we have proposed a penalized maximum-likelihood (PML) image reconstruction method, where the image reconstruction is performed with a regularization term that penalizes the dissimilarity between the target image and a prior image from a whole-body PET scan. Patient movement between the whole-body PET scan and two-panel PET scan can cause misalignment which may lead to artifacts in the PML reconstruction. In this work, we conduct a computer simulation study of the impact of the misalignment on the PML image reconstruction by translating the whole-body prior image by different distances. Results show that the PML reconstruction with misalignment being 1 mm is visually the same as the PML reconstruction with aligned whole-body prior image. The PML reconstruction with misalignment being 3 mm or 5 mm has artifacts, where the hot sphere is shifted and elongated. Correction for the misalignment caused by patient movement is important for artifact-free PML reconstruction from the two-panel dedicated PET scan.

Keywords: Misalignment, PET imaging, Patient movement, PML image reconstruction
Poster panel: 205

Poster Number:

Impact of Timing Resolution on DIRECT TOF PET Reconstruction with Sparse Detectors (#1593)

J. Li1, 2, Y. Lv2, J. Zhao1

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


TOF (Time-Of-Flight) information gives an accurate estimation of the point of annihilation, which can provide improved contrast-versus-noise and bias-versus-variance trade-offs, and faster convergence in iterative reconstruction. Moreover, additional (and redundant) information in the TOF data enables formulation of new reconstruction algorithms and data processing procedures.  In this work, we investigate the effect of improved TOF resolution on TOF-PET reconstruction for the PET systems with dense detectors and sparse detectors. An accelerated approach is proposed to implement forward- and backward-projection by taking advantage of sparsity of TOF data in DIRECT data framework, and it is demonstrated to be much faster than routine FFT-based approaches. Our experiments indicated that sparse detectors with a high timing resolution can yield image quality comparable to (or better than) that obtained from dense detectors with a low timing resolution.

Keywords: TOF, DIRECT, sparse detector
Poster panel: 208

Poster Number:

Validation of SPECT-CT image reconstruction for the Mediso AnyScan SCP scanner in STIR (#1661)

D. Deidda1, 2, B. Thomas2, K. Ferreira1, W. Heetun1, A. Forgács3, B. F. Hutton2, K. Thielemans2, A. Robinson1

1 National Physical Laboratory, Nuclear Medicine, Medical Radiation Physics, Teddington, United Kingdom
2 University College of London, Institute of Nuclear Medicine, London, United Kingdom
3 ScanoMed Ltd, Debrecen, Hungary


Single photon emission tomography (SPECT) is widely used in clinical practice for a large number of diagnostic applications. However, different hospitals might use different scanners containing different hardware and software technologies, therefore  making reproducibility of results a hard task. The aim of this work is to establish the basis for a tool able to include different clinical scanner models to facilitate inter-comparison of  activity measurements across different institutions. We used the open source software for tomographic image reconstruction (STIR) and implemented functionalities to read, process and reconstruct the data from the Mediso AnyScan SCP scanner. In particular,  the triple energy window method for scatter estimation,  and a method to re-scale the CT Hounsfield units in attenuation coefficient units (cm-1) were implemented and tested.   3D printed phantom data, with organ inserts filled with 177Lu, and a 99mTc clinical bone study were reconstructed using the implemented corrections and different anatomically-guided algorithms. The effect of the aforementioned correction was studied using ROI analysis and line profiles, whereas visual comparison was carried out between the reconstructed images with the vendor software and with STIR. Moreover, we demonstrated the feasibility of SPECT image reconstruction using the CT and iterative SPECT image estimates as prior information. Finally, a run time performance study showed that, when using multiple cores,  an acceleration of  a factor 2.7 is achieved for OSEM and around 2 for the other algorithms which involved more image-based operations. Other functionalities, such as list mode processing and reconstruction, are work in progress and  others scanners will be included in the future.

Keywords: SPECT, software, image reconstruction, SPECT-CT, 3D printing
Poster panel: 211

Poster Number:

Spatiotemporal Median Root Prior in MAP PET Image Reconstruction for Dynamic Neuroimaging (#1690)

J. J. Scheins1, E. Rota Kops1, L. Tellman1, C. Lerche1, N. J. Shah1, 2

1 Forschungszentrum Juelich GmbH, Institute of Neuroscience and Medicine 4 (INM-4), Juelich, Germany
2 RWTH Aachen University, Dep. of Neurology, Aachen, Germany


Dynamic neuroimaging using PET usually suffers from noise which increases with increasing temporal resolution, i.e. because of shorter frame length and lower count statistics. Therefore, efficient noise suppression techniques are highly desirable. In this context, MAP image reconstruction in combination with the robust Median Root Prior (MRP) is a practical option. We have now extended the conventional 3D MRP, which considers only the spatial domain, by additionally considering the time domain. This is motivated by the fact that most time activity curves (TAC) are expected to be smooth without discontinuities. Thus, the proposed spatiotemporal MRP realises a more efficient regularisation than 3D MRP with better noise suppression performance as demonstrated by brain tumour kinetics using the amino acid PET tracer O-(2-[18F]fluoroethyl)-L-tyrosine (FET).

Keywords: PET, MAP reconstruction, 4D, Median Root Prior
Poster panel: 214

Poster Number:

Accelerating primal-dual algorithm for non-convex optimization problems in CT reconstruction using a non-diagonal step-preconditioner (#1847)

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

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


In this work, we propose a non-diagonal preconditioning matrix for the primal-variable update step in the Chambolle-Pock (CP) or non-convex Chambolle-Pock (ncCP) algorithms, to accelerate their convergence rate. The non-diagonal step-preconditioner (NDPC) is based on the generalization of the proximal point algorithm, which allows for a matrix-mapping step, to the primal-dual algorithm. The NDPC used in this work is motivated by a saddle-point example and approximated with a truncated series for matrix inverse. Using the non-linear partial volume (NLPV) problem as an example, which features a non-linear data model, we have applied the NDPC to the CP and ncCP algorithms for TV-constrained least-square-minimization problems, and observed faster convergence of the data RMSE by using NDPC.

Keywords: first-order primal-dual, CT, reconstruction, non-diagonal step-preconditioner, non-convex
Poster panel: 217

Poster Number:

Scattering Noise Model Based EM Iterative Algorithm for XFCT Image Reconstruction (#2043)

S. Zhang1, L. Li1, Z. Chen1

1 Tsinghua, Engineering Physics, Beijing, China


X-ray fluorescence computed tomography (XFCT), which detects the x-ray fluorescence (XRF) photons emitted from the target fluorescence probes, is a high sensitivity imaging technique of High-Z elements (such as gadolinium (Gd) or gold (Au)) in the object. However, x-ray fluorescence signal is always affected by polychromatic Compton scatter background because recent benchtop XFCT devices use conventional x-ray tubes to stimulate fluorescent photons. As the statistical noise caused by scattered photons is difficult to be removed and will contaminate the XRF signal, a noise model based reconstruction algorithm might be necessary for XFCT image reconstruction. In this work, we presented an EM iteration algorithm for benchtop XFCT image reconstruction based on Poisson noise model. The estimation of the scatter background and the XRF signal was updated alternately during each iteration. Numerical simulations show that the contrast of the target element in the XFCT image significantly improves when the statistical noise of scattered photons is taken into consideration.

Keywords: EM iterative algorithm, Image reconstruction, XFCT, scattering
Poster panel: 220

Poster Number:

4D CBCT reconstruction with TV regularization on a dynamic software phantom (#2082)

R. Heylen1, G. Schramm1, J. Nuyts1

1 KULeuven, Department of Imaging and pathology, Leuven, Belgium

This work was done under the NEXIS project, that has received funding from the European Union's Horizon 2020 Research and Innovations Programme (Grant Agreement #780026)


In 2011 a primal-dual optimization algorithm was proposed by Chambolle and Pock. This algorithm is well fit for 4D CBCT reconstruction with spatial and temporal total variation (TV) regularization, and several authors recently used this approach. In this paper, we also implement a version of this algorithm, and assess its capabilities for dual-energy dynamic angiography of the brain on a software phantom. The phantom is created by adding an artificial vasculature, generated with constrained constructive optimization, to the brainweb phantom, and by calculating blood flow dynamics through this vasculature based on hydrodynamic equations. A dual-energy CBCT acquisition is simulated, split into water and iodine components, and a 4D iodine image is reconstructed. This suggests the viability of 4D CBCT techniques based on TV for dual-energy dynamic brain angiography.

Keywords: CT reconstruction, software phantom
Poster panel: 223

Poster Number:

Teacher-student Network for CT image Reconstruction via Meta-learning Strategy (#2126)

M. Zhu1, 2, S. Li1, 2, D. Li1, 2, Q. Gao1, 2, D. Zeng3, 1, Z. Bian1, 2, J. Huang1, 2, J. Ma1, 2

1 Southern Medical University, School of Biomedical Engineering, Guangzhou, China
2 Southern Medical University, Guangzhou Key Laboratory of Medical Radiation Imaging and Detection Technology, Guangzhou, China
3 South China University of Technology, College of Automation Science and Engineering, Guangzhou, China

This work was supported in part by the National Natural Science Foundation of China under Grants 61701217, 81701690, 61571214 and U1708261, the Science and Technology Program of Guangzhou, China under Grant 201705030009, the Science and Technology Program of Guangdong, China under Grant 2015B020233008.


Deep neural networks (DNN) have been widely used in computed tomography (CT) imaging, with promising performance. Meanwhile, most of them are supervised learning strategies, and their performance highly depends on the amount of the pre-collected training samples. In the training, the high-dose CT images are usually chosen as labels, but this data is sometimes hard to be collected due to the cancer-risk of high-dose CT scanning. Instead, the unlabeled low-dose CT images are easy to access, but they fail to incorporate a large amount of latent information contained into network training. To address these two issues, in this work, we present a couple teacher-student DNN strategy for low-dose CT image reconstruction via meta-learning strategy. Specifically, this strategy mainly consists of two network, i.e., teacher network and student network. In the teacher network training, only a small amount of samples with high-quality labels (low-dose/high-dose CT image pairs) are included. Then, the unlabeled low-dose CT data are enrolled into this trained teacher network for processing to obtain the temporary high-quality ones. Finally, the temporary high-quality and another a small number of pre-collected samples with high-quality labels are combined into the student network training. For simplicity, the proposed method is terms as “metaCT”, which is similar to the meta-learning strategy containing teacher network and student network. Moreover, the present metaCT is fully flexible to adopt the existing DNN-based CT image reconstruction model as the teacher/student network, while the ResNet framework was used in the two network in our work. Experiments on the Mayo clinic dataset demonstrate that the present metaCT method is effective in low-dose CT image reconstruction with a small amount of labeled data and a large amount of unlabeled data.

Keywords: CT reconstruction, deep neural networks, meta-learning, teacher network, student network
Poster panel: 226

Poster Number:

Optimization of a customized Simultaneous Algebraic Reconstruction Technique algorithm for breast CT (#2191)

S. Donato1, 2, L. Brombal1, 2, F. Arfelli1, 2, V. Fanti3, 5, R. Longo1, 2, P. Oliva4, 5, L. Rigon1, 2, B. Golosio3, 5

1 University of Trieste, Physics, Trieste, Italy
2 INFN division of Trieste, INFN, Trieste, Italy
3 University of Cagliari, Physics, Monserrato, Italy
4 University of Sassari, Chemistry and Pharmacy, Sassari, Italy
5 INFN division of Cagliari, INFN, Monserrato, Italy


Iterative CT reconstruction algorithms coupled with edge-preserving filters are attracting a growing interest in the field of biomedical X-ray imaging. In many cases the application of such algorithms results in an improved reconstruction quality when compared with filtered back-projection (FBP). Iterative algorithms commonly entail a decrease of image noise or, equivalently, an increase of contrast-to-noise ratio, while preserving image detail. Conversely, they modify the shape of noise power spectrum, producing a shift towards lower spatial frequencies with respect to FBP. This can result in a “patchy” or “waxy” appearance of the reconstructed images. Changes in image texture affect radiologists’ perception of image quality, possibly influencing their willingness to use an iterative algorithm in clinical practice. In this work we present a GPU implementation of a simultaneous algebraic reconstruction technique algorithm, combined with a bilateral regularization filter, and we discuss the optimization of the algorithm’s parameters in terms of noise texture, selecting those parameters that preserve its “natural” appearance. We evaluated the performances of the algorithm, compared to FBP, both on a test phantom and on a surgical mastectomy sample by using contrast-to-noise ratio and spatial resolution metrics. Samples were imaged at the SYRMEP beamline of the Elettra synchrotron facility with a monochromatic beam (32 keV) in propagation-based phase-contrast configuration, delivering a clinically-compatible radiation dose of 5 mGy and using a large-area CdTe photon-counting detector. Results show, in the specific application on breast specimens, that the implemented algorithm can be tuned to preserve the noise texture and spatial resolution observed in FBP reconstructions, while improving contrast-to-noise ratio up to 30%.

Keywords: Breast CT, Iterative algorithms, Noise texture
Poster panel: 229

Poster Number:

Evaluation of Data-Quality-Adaptive Image Domain PSF Modeling for PET Image Reconstruction (#2260)

L. Yang1, W. Qi1, C. Chan1, E. Asma1

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


We developed a data-quality based image domain PSF modeling approach in order to improve contrast recovery while avoiding visible image artifacts such as ringing. We developed rules for kernel widths such that more aggressive, broader kernels were used for noisier data and narrower kernels were used for higher quality data. We analyzed the ROI quantitation performance of this approach and compared it to the standard fixed-kernel-width PSF modeling.  We analyzed five whole-body patient scans containing both real and inserted lesions. The quantitative metrics were the contrast recovery coefficient (CRC) for inserted lesions and the mean lesion standard uptake value (SUV) for real lesions versus liver background variability. The results show that the proposed data-quality adaptive PSF modeling approach can achieve better quantification performance compared to standard PSF modeling without enhancing any ringing artifacts.

Keywords: PSF modeling, PET reconstruction
Poster panel: 232

Poster Number:

Estimation of Image Parameters Directly from Sinograms using Neural Networks (#2331)

H. Chang1, U. Sreshtha2, Y. Seo2, G. T. Gullberg2, D. Mitra1

1 Florida Institute of Technology, Computer Science, Melbourne, Florida, United States of America
2 University of California, San Francisco, Radiology, San Francisco, California, United States of America


Neural networks (NNs) have been known to be suitable for various mathematical transformations. With the advent of GPU-based implementations of NNs, they are making a significant impact on the application of artificial intelligence in medicine. In this project we investigate the NN’s usefulness in relation to the reconstruction of Radon projection data. Medical image reconstruction is a type of estimation problem. For example, the estimation of emission counts or absorption coefficients from noisy Radon projections of an image, also known as a sinogram. We investigate with a simple simulation how different parameters of an image can be recovered directly from sinograms with NN models. We have estimated object locations in 2D, radii for circular objects, constant attenuation coefficients, and shapes of complex 2D objects when they are used in creating sinograms. We investigated different NN architectures including a convolutional neural network.

Keywords: deep learning, parameter estimation, radon transformation, image reconstruction, tomography
Poster panel: 235

Poster Number:

Dynamic Tomographic Reconstruction of Hepatobiliary scintigraphy from dynamic planar imaging followed by SPECT (#2366)

L. Presotto1, A. Savi1

1 IRCCS San Raffaele Scientific Institute, Nuclear medicine unit, Milano, Italy


Hepatobiliary scintigraphy allows the estimation of liver function but only for the whole liver and not for individual regions. A workaround for this consists in rotating the camera to the angle where maximum separation between the lobes is achieved, however, this does not fully provide the activity distribution. In this work, we introduce an algorithm to reconstruct a full 4D tomographic image by using the same conventional planar dynamic imaging sequence, followed by a tomographic acquisition. The algorithm is based on a dictionary learning approach, where multiple L1 and L2 regularization terms are applied both on the bases and on the coefficients. The novelty of this approach is in that we fully model the Poisson statistics of the problem and of photon attenuation, which improves image quality compared to previous approaches but makes convergence extremely slow. This was solved using an optimal diagonal preconditioner and Nesterov acceleration. In this way, some depth information can also be achieved by the projection consistency when attenuation correction is used. We report the results obtained on a patient case. The image quality improves better the closer a frame is to the tomographic acquisition. VOI analysis show activity trends similar to that of planar acquisitions. Visually the images produced seem to allow a much better discrimination of the different regions of the liver. The problem of validation remains open.

Keywords: dynamic imaging, undersampled reconstruction, spect
Poster panel: 238

Poster Number:

4D Reconstruction with Projection and Image Domain Motion Estimation (#2452)

S. Zhou1, Y. Chi1, J. Wang2, M. Jin1

1 University of Texas at Arlington, Department of Physics, Arlington, Texas, United States of America
2 University of Texas Southwestern Medical Center, Department of Radiation Oncology, Dallas, Texas, United States of America


To better reconstruct 4D cone-beam computed tomography (CBCT) images, a general simultaneous motion estimation and image reconstruction (G-SMEIR) method is proposed to mitigate the local optimum trapping problem of the original SMEIR method. In addition to the projection domain motion estimation used in SMEIR, G-SMEIR includes an image domain motion estimation in the iteration to achieve better 4D reconstruction. To improve computational efficiency, the computationally intensive image domain motion estimation is alleviated by parallel computing using graphics processing units (GPUs). The proposed G-SMEIR method is tested using a CBCT simulation study of 4D NCAT phantom at different noise levels and compared with 3D total variation-based reconstruction (3D TV) and SMEIR. G-SMEIR performed similarly at the regular and low doses. The root mean square error of G-SMIER is improved more than 60% over 3D TV and up to 17% over SMEIR. The structural similarity indices for a representative phase are 0.6418 (3D TV), 0.8893 (SMEIR), and 0.9206 (G-SMEIR).  GPU computing shortens computational time of image domain motion estimation from about 15 minutes (CPU) to about 40 seconds for each pairs of 3D images. The simulation results demonstrate that G-SMEIR yields good image quality at different noise levels in a reasonable time. Further improvement of motion estimation algorithm and full parallelization of G-SMEIR will be conducted and tested on patient data.

Keywords: 4D image registration, Cone-beam computed tomography (CBCT), Motion estimation, GPU computing
Poster panel: 241

Poster Number:

On C-arm CT imaging with the extended line-ellipse-line trajectory   (#2608)

Z. Guo1, 2, G. Lauritsch3, A. Maier2, P. Kugler3, M. Islam4, F. Vogt3, F. Noo1

1 University of Utah, Salt Lake City, Utah, United States of America
2 University of Erlangen-Nuremberg, Erlangen, Bavaria, Germany
3 Siemens Healthcare GmbH, Forchheim, Bavaria, Germany
4 SYNG4 GmbH, Forchheim, Bavaria, Germany


Cone-beam (CB) imaging in interventional radiology, which we refer to as C-arm CT, is a valuable tool that could use further improvements to support growing clinical needs. Typically, a circular short-scan is used for data acquisition, which leads to unavoidable CB artifacts and limits the axial coverage to that defined by the axial extent of the detector. For robotic C-arm systems, it may be possible to overcome these two limitations using alternative data acquisition geometries. Based on previous work by our group and others, the extended line-ellipse-line (LEL) trajectory is emerging as an attractive candidate for this purpose. This trajectory is suitable to achieve efficient reconstruction with no CB artifacts, as well as to image any desired volume length in the axial direction. This second feature can be used to better cover long anatomical sites, such as the aorta or the spine; or to allow imaging with less scatter without sacrificing axial coverage, through axial beam collimation. Recently, we showed that the extended LEL trajectory can be  implemented on a state-of-the-art robotic C-arm system with good geometrical fidelity and reproducibility. Here, we investigate the performance of this first implementation in terms of CB artifacts relative to the use of a circular short-scan. For this purpose, we compare iterative reconstructions of an anthropomorphic head phantom. Our results are encouraging in that they show the expected elimination of CB artifacts as well as the increase in axial coverage.

Keywords: C-arm CT, cone-beam artifact, line-ellipse-line trajectory, iterative reconstruction
Poster panel: 244

Poster Number:

Accelerated Regularised List-Mode PET Reconstruction Using Subset Relaxation (#2663)

M. G. Spangler-Bickell1, T. Deller2, F. Jansen2, K. Wangerin2

1 IRCCS Ospedale San Raffaele, Nuclear Medicine Unit, Milan, Italy
2 GE Healthcare, PET/MR Engineering, Waukesha, Wisconsin, United States of America


PET image reconstruction using regularisation algorithms can provide good image quality and ensure convergence with suitable parameter selections, however they usually require many iterations to do so. A list-mode form of the BSREM regularised algorithm for PET image reconstruction is presented, with an acceleration technique whereby the number of list-mode events used in each subset is increased with increasing iteration number. This allows for quick convergence in early iterations, and avoids noise propagation from ``small'' subsets as well as limit cycles in later iterations.

Keywords: PET image reconstruction, Reconstruction acceleration, Regularized reconstruction, Subset relaxation, List-mode reconstruction
Poster panel: 247

Poster Number:

A Deep Learning Method for Image based Anti-aliasing in CT Scanners with Single Focal Spot Acquisition (#1105)

Y. Bao1, Z. Stanislav1, G. T. Quan1

1 United Imaging Healthcare Co., Ltd, CT&MI RPA, shanghai, China


This paper describes a deep learning method that provides an improved image quality for single focal spot CT scan. Based on 30 clinical head datasets of flying focal spot images and the corresponding single focal spot images, an experiment of convolution neural network is presented. The generated anti-aliasing images reduce streak artifacts and noise granularity, and improve the contrast of bony structure, which are potential to meet with the high diagnostic criteria in clinical applications while keep the CT system stability, compared to flying focal spot scan. Further, it can be extended into more widely applications such as inner ear and extremities.

Keywords: deep learning, ct, flying focal spot
Poster panel: 250

Poster Number:

A FCN-based Unsupervised Learning Model for Deformable Chest CT Image Registration (#1374)

Q. Fang1, 2, X. Gu1, 2, J. Yan2, J. Zhao1, Q. Li3

1 Shanghai Jiao Tong University, School of Biomedical Engineering, Shanghai, China
2 Shanghai United Imaging Healthcare Co., Ltd, Corporate Research Center, Shanghai, China
3 Huazhong University of Science and Technology, Wuhan National Laboratory for Optoelectronics, Wuhan, China


Image registration is a fundamental technique for many automatic medical image analysis tasks, but it can be time-consuming, especially for deformable three-dimensional image registration. In this paper we propose a fast unsupervised learning method for deformable image registration using a fully convolutional network (FCN). The network directly learns to estimate a dense displacement vector field (DVF) from a pair of input images. A spatial transform layer then uses the DVF to warp the moving image to the fixed image. Different from supervised learning based image registration methods, the network is trained by maximization of a similarity metric between the fixed image and the warped moving image. Thus training does not require supervised information such as manually annotated or synthetic ground truth. We evaluate the proposed model on publicly available datasets of inspiration-expiration chest CT image pairs. The results demonstrate that our method achieved an accuracy comparable to that of a state-of-the-art conventional image registration method, but executed orders of magnitude faster.

Keywords: Chest CT, deformable image registration, fully convolutional network, unsupervised learning
Poster panel: 253

Poster Number:

Fast Estimation of Compartmental Model Parameters using Model-based Input Function (#1580)

Y. Zhao1, T. Feng2, Y. Lv1, Y. Ding1, Y. Dong1, D. Hu1

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


Parametric imaging has been shown to provide better quantitative results and other clinical values. With the recently-developed high sensitivity total-body PET scanner, total-body parametric imaging using the compartmental model with high temporal resolution becomes possible. The input function acquired using conventional approach consists of discrete data points. Interpolations are often required when there is a time mismatch between the tissue time-activity-curve (TAC) and the input function, which could increase the complexity, and calculation time of the parameter estimation, which becomes an importance factor due to the much increased data acquired by the total-body PET. Current PET scan usually uses a bolus injection which makes the input function a single-peak shape. It can be assumed that before the peak, the input function is a mono-increasing function and a mono-decreasing function after the peak. In this work, the input function was fitted as piecewise functions consist of linear and exponential functions. Two-tissue irreversible compartmental model was used for this study. The Levenberg-Marquardt method was used for parameter estimation. Clinical dynamic data acquired from the total-body PET scanner as well as simulated TACs was included for method validation. The TAC was simulated using the 2-tissue compartment model with parameters reported in the literature. Different noise levels were simulated to test the robustness of the method. Both the model-based input function and conventional discrete input function were employed for method validation. The estimation error of the kinetic parameters from the model-based input function method was less than 5% when k4<0.01 and TAC SNR>10dB, when the computing speed was 20 times increased. The results showed that this method can effectively shorten the calculation time by a factor of 20 with comparable estimation accuracy compared with non-model-based input functions.

Keywords: compartmental model, fast estimation, input function
Poster panel: 256

Poster Number:

Slice interpolation of medical images via fuzzy system and radial basis function neural networks  (#1632)

Z. Chao1, H. - J. Kim1

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


Volume data composed of complete slice images play an indispensable role in medical diagnosis. However, system or human factors often lead to the loss of slice images. Based on this, compensation for missing data has become a hot research direction. Interpolation methods are a powerful tool to compensate for missing slice data. In recent years, various interpolation algorithms have been proposed to solve this problem. Although these methods are effective, the interpolated images still have some shortcomings. In this study, we propose a new method based on an enhanced fuzzy radial basis function neural network to improve the performance of interpolation method. An enhanced fuzzy radial basis function neural network is established by including six input neurons and one output neuron. Accordingly, we use two normal pending images that need to be interpolated as input. The output data obtained is used as the actual output data for neural network training. Herein, we use the-state-of-the-art interpolation method to obtain an interpolated image that can be taken as ideal output data for training a neural network system. To train and update the neural network, a hybrid of the gravitational search algorithm (GSA) and error backpropagation algorithm (EBPA) are proposed. Finally, we obtain the optimal interpolated images using the evolved neural network. In examining a group of brain MRI images, the proposed method outperforms other method with regard to subjective observation and objective evaluation.

Keywords: Missing slice images, Slice interpolation method, Enhanced radial basis function neural network, Hybrid of gravity search algorithm and error backpropagation algorithm
Poster panel: 259

Poster Number:

An Evaluation of the Quantitative Advantages of Combined Kinetic Analysis of Multiple Injection Dynamic PET Scans (#1550)

F. Gu1, F. O'Sullivan1, M. Muzi2, D. A. Mankoff3

1 University College Cork, Cork, Ireland
2 University of Washington, Seattle, United States of America
3 University of Pennsylvania, Philadelphia, United States of America


The multiple injection dynamic Positron Emission Tomography (PET) scanning is sometimes used in the clinical management of certain groups of cancer patients and in medical research. The analysis of such studies can be approached in one of two ways: analyze individual injections separately to recover tracer kinetic information, or concatenate data from separate injections and carry out a combined analysis. Separate analysis offers some simplicity but may not be as efficient statistically. We evaluate these approaches in a setting where the mapping of kinetic information is based on mixture analysis. The mixture technique is readily implemented in a separated or combined analysis mode. Our work reports on set of 1D simulation studies matched to the mathematical complexity of PET. These simulations are largely guided by experience with breast cancer flow-metabolism mismatch studies using O-15 water (H2O) and F-18 Fluorodeoxyglucose (FDG). An efficient implementation in the R (an open-source environment) is used to implement simulations. The simulations evaluate Mean Square Error (MSE) characteristics, for separate and combined analysis, both as a function of overall dose and relative dose (H2O/FDG). The relationship between MSE characteristics of the underlying source distribution and local kinetic parameter MSE is described. Regardless of dose, the combined analysis is found to reduce MSE by between 9% and 15%. Some theoretical analysis is developed to further support this finding.

Keywords: Multiple Injections, Combined Kinetic Modelling, Simulation, Mixture Analysis
Poster panel: 262

Poster Number:

An Adaptive Boosting Strategy for GLCM-CNN Model in Differentiating the Malignant from Benign Polyps (#1860)

S. Zhang1, Z. Liang1, W. Cao1, Y. Gao1

1 Stony Brook University, Radiology, Stony Brook, New York, United States of America


In recent years, deep learning such as Convolutional Neural Network (CNN) has demonstrated its superior in the field of image classification.  However, in the medical imaging field, it still faces great challenges for tumor classification in computer-aided diagnosis due to uncertainties of lesions including their size, scaling factor, rotation, shapes, etc.  Hence, texture pattern is an option to be fed into the CNN model and gray level co-occurrence matrix (GLCM) can be chosen as the texture pattern for its several good properties such as uniform size, shape invariance, posture robustness, scaling invariance.  For a 3D volume data, 13 independent GLCMs could be extracted according to 13 digital distinct directions.  The sampling along the 13 directions has different sampling displacement across the 3D voxel array.  Considering this multi-scale sampling is hypothesized to be optimal compared to ignoring the multi-scale nature.  In this work, we proposed an adaptive boosting learning (ABL) to test the hypothesis.  By the multi-scale nature, the 13 directions are grouped as three, and the ABL is applied to learn the best classification performance from the 3 subgroups.  By comparing our ABL with and without the multi-scale nature, our experiments showed a gain of 0.5~3.7% in terms of AUC (area under the ROC curve) for polyp classification, reaching AUC of 91.13% for a small database of 63 samples with pathological reports.  The gains indicate that the GLCM-CNN model with ABL on the textures can mitigate the challenges for classification of limited number of tumors in medical imaging field.

Keywords: Polyp Classification, CNN, GLCM, Adaptive Boosting Strategy
Poster panel: 265

Poster Number:

Identifying Demyelinating and Ischemia brain diseases through magnetic resonance images processing (#2102)

D. P. Castillo1, 2, R. Samaniego3, Y. Jiménez1, J. M. Álvarez-Gómez2, L. Cuenca1, O. Vivanco4, M. J. Rodríguez-Álvarez2

1 Universidad Técnica Particular de Loja, Chemestry and Exact Sciences, Loja, Ecuador
2 Universitat Politècnica de València, Instituto de Instrumentación para Imagen Molecular (i3M), Valencia, Spain
3 Universidad Técnica Particular de Loja, Hospital, Loja, Ecuador
4 Universidad Técnica Particular de Loja, Biological Sciences Department, Loja, Ecuador


Brain Magnetic Resonance Images are a very useful tool for the diagnosis of brain diseases and analyze brain changes. The appropriate processing (neuroimaging) can help to identify, measure and classify different lesions or abnormalities. The principal aim of this project is to develop an algorithm that can identify and differentiate the ischemic disease than the demyelinating disease in the brain through the processing of magnetic resonance images. The damage and deterioration of the myelin layer of nerve fibers (brain demyelination) is the cause of pathologies like multiple sclerosis. Ischemic stroke is produced by the interruption of the blood supply to the brain. The dataset used was composed by images T1, T2 and FLAIR modalities of 90 patients from the hospital. For the segmentation of the features, the identification and the classification of the lesions has used the methods of Discrete Wavelet Transform (DWT), principal component analysis (PCA) and support vector machine (SVM). The results present to 80% of accuracy to identify and differentiate the diseases.

Keywords: brain disease, image processing, MRI, demyelinating, ischemia disease
Poster panel: 268

Poster Number:

Effect of Image Noise on Registration in PET Brain Imaging (#2153)

M. G. Spangler-Bickell1, T. Deller2, S. A. Hurley3, A. B. McMillan4, V. Bettinardi1, F. Jansen2, K. Wangerin2

1 IRCCS Ospedale San Raffaele, Nuclear Medicine Unit, Milan, Italy
2 GE Healthcare, PET/MR Engineering, Waukesha, Wisconsin, United States of America
3 University of Wisconsin - Madison, Radiology, Madison, Wisconsin, United States of America
4 University of Wisconsin, Madison, Wisconsin, United States of America


Rigid image registration techniques can be used in PET imaging to align frames acquired during a brain scan where motion may or may not have been present during the scan. The subsequent estimated motion parameters can be used to correct for the motion. Shorter frames provide better temporal resolution but suffer from increased noise, which may degrade the quality of the image registration. The relationship between the image noise (by proxy of the frame count level) and the accuracy of the image registration is investigated. For FDG brain studies, a trues count level of 200x103 (usually approximately 1 sec frame duration) was found to provide adequate image quality for accurate registration.

Keywords: PET reconstruction, Image registration, Motion correction
Poster panel: 271

Poster Number:

Motion Compensation in dual energy X-ray imaging based on Deformable Image Registration (#2309)

P. Palaniappan1, P. Steininger2, I. M. Messner2, H. Deutschmann2, K. Parodi1, G. Landry1, M. Riboldi1

1 Ludwig-Maximilians-University, Faculty of Physics/Chair of Medical Physics, Munich, Bavaria, Germany
2 MedPhoton GmbH, Salzburg, Austria


The aim of the study was to explore the use of dual energy X-ray imaging for motion management in external beam radiotherapy. We specifically developed a method to increase the quality of chest radiography in dual energy acquisition mode, when fast kVp switching is used to acquire sequential frames at low and high energy. Dual-energy imaging offers advantages such as higher spectral separation with enhanced contrast and detection quantum efficiency, but resulting images are affected by artifacts caused by motion of the anatomical structures in between sequential frames. A quality improvement method was designed and tested, relying on deformable image registration to align the high energy and low energy images from the detector. The non-rigid multi resolution deformable image registration algorithm reported here operates in multi-level, starting from the lower resolution towards a finer resolution scale. In order to separate the bony anatomy structures from the soft tissues, the registered images were processed through a dual energy weighted log subtraction algorithm. For this study, the data were acquired using the ImagingRing System that supports fast kV switching. The potential for motion compensation and weighted log subtraction are clearly characterized in the soft tissue images. The accuracy of the registration was quantified using the absolute mutual information and around 5 percent improvement was seen in the registered image compared to the image with motion artifacts. This resulted in visually improved soft tissue images, especially in terms of rib removal. Further studies will extend the initial results to improve the computation latency and accuracy. The implemented algorithm will be applied for lung tumor tracking, relying on interleaved registration of high energy and low energy images to provide motion compensation at the imaging frame rate.

Keywords: Dual-energy imaging, Deformable image registration, Motion artifacts
Poster panel: 274

Poster Number:

4D respiratory motion synchronized image synthesis from static CT images using GANs (#2389)

V. Jaouen1, D. Visvikis1

1 LaTIM, Université de Bretagne Occidentale, Inserm, Brest, France


Respiratory motion is a major limiting factor in computed tomography and positron emission tomography, especially in the thorax and in the abdomen. While four-dimensional computed tomography (4DCT) acquisitions may yield some degree of motion-awareness, it is at the cost of increased radiation dose delivered to the patient. In this paper, we are interested in synthesizing realistic phase-gated 4DCT data from single-helical CT images that could serve as surrogates for real 4DCT acquisitions. More precisely, we study the feasability of respiratory motion synthesis using deep conditional generative adversarial learning networks trained on real 4DCT data. The proposed network architecture is an ensemble of parallel image-to-image translation networks, each synthesizing one respiratory phase-gated frame of the 4DCT acquisition. We show that realistic breathing dynamics can be obtained from a small set of training 4DCT data. Although not clinically meaningful, these results constitute a first step among ongoing developments in our laboratory towards the synthesis of patient-specific pseudo-4DCT acquisitions from 3DCT images, where external breathing tracking system would further condition motion synthesis.

Keywords: Image standardization, image synthesis, machine learning, generative adversarial networks
Poster panel: 277

Poster Number:

Application of Mixture Analysis Tools from PET for Quantitation of Local Kinetics in DCE-MRI (#2469)

Y. Hu1, J. Huang1, M. Muzi2, F. O'Sullivan1

1 University College Cork, Department of Statistics, Cork, Ireland
2 University of Washington, Department of Radiology, Seattle, Washington, United States of America


Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) studies use MRI measured T1-signal changes induced by an intravenously injected paramagnetic substrate to evaluate vascularity characteristics (flow and perfusion) of local tissue. The standard analysis begins by transformation of the raw signal to obtain the time-course of the concentration of the contrast agent. Similar to classical indicator dilutions, the transformed time course is represented as a convolution between the arterial input function and the tissue impulse response. Parametric 2-compartment representation for the impulse response (the so-called Toffs model) is frequently used for quantitative kinetic analysis. Because of the conceptual framework has much in common with PET studies, we examine the adaptation of a mixture analysis methodology, developed for kinetic mapping of dynamic PET studies, to DCE-MR quantitation. In the approach, mixing coefficients are based on piecewise constant approximations whose elements are 3-D indicators arising from domain segmentation. The optimized mixture model is formulated in terms of a regularized estimation process that gives the ability to locally adjust model smoothness. This process is adapted using generalized cross-validation. The mixture modelling gives the ability to recover voxel-level analysis of the tissue residue and to map by a linear process the associated tracer kinetic parameters. We review this approach emphasizing links with more familiar 1-C and 2-C compartmental analysis of PET kinetics. We illustrate the approach by application to cerebral DCE-MRI study in a brain tumor patient, post-surgery. Results are encouraging. Model diagnostics highlight opportunities to optimize the analysis. Unlike PET, where standardized model residuals are well-described by Gaussian statistics, DCE-MRI indicates more heavy-tailed distributional characteristics and the need to use more robust fitting procedures to improve the efficiency of the approach.

Keywords: DCE-MRI, Mixture Models, Segmentation, Non-parametric Residues, Diagnostic Analysis
Poster panel: 280

Poster Number:

Direct Estimation of Neurotransmitter Activation Parameters in Dynamic PET Using Regression Neural Networks (#2610)

Y. Hu1, 3, G. I. Angelis1, 2, P. L. Kench1, Y. Liu3, 4, T. Ma3, 4, S. R. Meikle1

1 University of Sydney, Brain and Mind Centre, Sydney, Australia
2 Royal North Shore Hospital, Radiation Oncology, Northern Sydney Cancer Centre, Sydney, Australia
3 Tsinghua University, Department of Engineering Physics, Beijing, China
4 Ministry of Education, Key Laboratory of Particle & Radiation Imaging, Beijing, China


Current pharmacokinetic models, such as the linear parametric neurotransmitter PET (lp-nTPET) model have been developed to detect and quantify transient changes in receptor occupancy caused by variations in the concentration of endogenous neurotransmitters. The lp-ntPET model offers reliable parameter estimates when applied at the region of interest level. However, it performs poorly when applied at the voxel level due to high noise level.
In this paper, we proposed a new method using machine learning to detect transient changes in neurotransmitter concentration in dynamic PET data. Activation onset time and response magnitude of neurotransmitter were directly estimated using convolution neural network instead of using lp-ntPET model. We used computer simulations to generate dynamic PET data, representing a [11C]raclopride study, with a known range of activation onset time and response magnitude, across a wide range of noise levels.
Our results showed that neural network had better quantitative performance in estimating activation onset time and response magnitude than conventional lp-ntPET model, especially at voxel level where noise is high. Work is currently in progress to apply this method to data generated from GATE simulations and real animal experiments with environmentally induced dopamine activations.

Keywords: machine learning, lp-ntPET model, dynamic PET imaging, neurotransmitter activation parameter
Poster panel: 283

Poster Number:

Distributed Time Alignment Using L1-Norm Optimization for Total-body PET (#1069)

X. Lyu1, X. Wang1, S. An1

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


The PET uses accurate timing information to pair two 511keV photons into a coincidence event based on the truth that the timestamp assigned to each detector is synchronized. In real-world system, there always exist time delays from clock skew, optical pathway length etc. With improvement of timing resolution, time alignment becomes increasingly important as the timing calibration could quickly become the most important error contributor. As the industry partner of the EXPLORER consortium, UIH has built the EXPLORER PET/CT scanner with a ring diameter of 78.6cm and 194cm length in axial. In previous work, we have introduced the time alignment method implicated to this total-body PET using a weighted linear least squares iteration approach which could be referred as L2-norm optimization. Although this method can calculate the time offset of each crystal with high accuracy, it requires plenty statistics to avoid influence of outliers. In this work, we extend the time alignment method using L1-norm minimization, which is robust to random presence, applicable to low statistics and suitable for the long axial and complex total-body PET system. Long axial FOV of this PET system means a wide coincidence window needed to accept very oblique LOR events. Furthermore, a much more wide coincidence window is required while initially time calibrating. We can do the time calibration using this method through a very fast data acquisition making use of the high sensitivity of the total-body PET and capability of handling low statistics. This new time alignment approach based on L1-norm minimization is suitable for this situation. We have tested the accuracy and converge of this method using simulating data and validated on the real data from EXPLORER PET/CT by observing the time resolution improvement and TOF image quality.

Keywords: EXPLORER PET/CT, L1-norm, time alignment
Poster panel: 286

Poster Number:

Emis2Trans: Attenuation Correction for Brain PET With Many Types of PET Ligands Using Convolutional Neural Networks (#1146)

F. Hashimoto1, M. Ito2, K. Ote1, T. Isobe1, H. Okada2, Y. Ouchi3

1 Hamamatsu Photonics K.K., Central Research Laboratory, Hamamatsu-City, Japan
2 Hamamatsu Photonics K.K., Global Strategic Challenge Center, Hamamatsu-City, Japan
3 Hamamatsu University School of Medicine, Department of Biofunctional Imaging, Preeminent Medical Photonics Education & Research Center, Hamamatsu-City, Japan


Attenuation correction is imperative in positron emission tomography (PET) measurements. A traditional transmission scan with an external positron line source has been replaced by X-ray CT or MRI. However, the scanner’s mechanics increases cost and these additional scans sometimes accompany image displacement which makes final images degraded in quality.  Therefore, in this study, to overcome these problems, we propose an attenuation correction framework that generates transmission images from uncorrected emission images for brain PET imaging using deep convolutional neural networks (CNN). The advantage of this study is that an estimated transmission image is applicable in PET measurements with various types of ligands. Using encoder–decoder-based CNNs, the transmission scans for attenuation correction are generated from the uncorrected emission scans. We used the brain PET datasets from 1030 patients scanned with the following PET ligands: 18F-FDG, 18F-BCPP-EF, 11C-Racropride, 11C-PIB , 11C-DPA-713, and 11C-PBB3. We randomly selected 20% of the datasets as the testing dataset, and used the remaining 80% to train the network. As a result, the CNN generated was less noisy and had more uniform transmission scans compared to the original transmission scans (using a 68Ge-68Ga rotation line source). The CNN showed the peak signal-to-noise ratio and structural similarity indices as being 31.5 ± 1.8 dB and 0.80 ± 0.02 in generated and original transmission images of many types of PET ligands, respectively. These results indicate that the proposed CNN-based framework allows accurate attenuation correction in brain PET system using different types of PET ligands.

Keywords: Attenuation correction, Convolutional neural networks, Deep learning, Positron emission tomography (PET)
Poster panel: 289

Poster Number:

2D Energy Histograms for Scatter Estimation in an SiPM PET Scanner (#1206)

J. J. Hamill1

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


Introduction. PET measurements include a background of scattered radiation which must be estimated as a part of image reconstruction. In principle the estimate can be based on a physics model of the 2D energy histograms, eliminating the need for single-scatter simulation (SSS) and making the scatter estimation independent of the mu map. Past attempts to do this have shown promise but were limited by the dynamic range of energy data, sometimes based on just one bit of energy digitization, and by statistical noise in the measurements. In this work, we use the 6-bit energy digitization available in Siemens Biograph Vision SiPM-based scanners, and a denoising technique proposed in 2006 by Popescu et al. We ask if the histograms can be suitably denoised by this approach.
Methods. List-mode data were acquired based on PET line sources and uniformly filled cylinders. Since scatter sinograms vary smoothly as a function of spatial coordinates, we treated large groups of “A” and “B” crystals as single detectors so that sinograms could be made with millions of coincidences per bin. Energy signals were quantified as 2D A-vs.-B histograms. We used Popescu’s MLEM method to fit the histograms in an expansion with 6 basis functions for A and 6 for B. From this we calculated the scatter fraction in individual sinogram bins. Then we reduced the counts below the levels expected in clinical scans, using a random number generator, and again calculated the scatter fraction.
Results. The scatter fraction was essentially constant as the counts per bin were reduced 10,000-fold.
Conclusion. De-noising by the MLEM method was successful. The energy-based approach to scatter estimation was stable at low count levels. The method’s accuracy has not yet been investigated. The use of energy-based scatter estimation in PET reconstruction, and the comparison to SSS, are a work in progress.

Keywords: PET, energy, scatter
Poster panel: 292

Poster Number:

A Look-up Table Based Approach to Estimate Crystal Singles Rates for High-Activity PET Studies (#1365)

M. Aykac1, V. Y. Panin1

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


It is essential to measure prompts and randoms coincidence counts accurately for better quantification of PET images. The most common method to measure the randoms coincidence is to use delayed circuitry in most of the commercial PET scanners. The scope of this study is to investigate an alternative method to estimate singles by using a look-up table approach to be used in randoms smoothing and potentially eliminate the need of measuring the delayed coincidences. The basic idea behind the look-up table based approach relies on how the detected counts are distributed across the detector at various activity levels. Proposed method is tested on a uniform water cylinder and a Torso phantom with cold arms. The overall difference in singles estimation between the proposed method and traditional method based on the measured randoms is estimated to be around 3.6% for uniform cylinder and 4.1% for Torso phantom. The average region-of-interest value in the background region in both phantoms is in very good agreement within 1%. Look-up table based approach for randoms smoothing looks very promising and it should provide some relief to the scanner electronics for high activity acquisitions.

Keywords: randoms smoothing, singles, PET, Positron emission tomography
Poster panel: 295

Poster Number:

Influence of Optical Flow parameters on PET motion correction accuracy: a phantom evaluation study (#1466)

S. Pösse1, F. Büther1, 2, M. Schäfers1, 2, K. P. Schäfers1

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


Optical Flow (OF) is an established method to estimate motion between different respiratory phases for PET motion correction. Different intrinsic parameters of the OF algorithm can result in different motion vector fields (MVF). In our previous study we developed an optimization of these parameters with patient data and received an optimal parameter set. On this basis this study focuses on the question whether the optimized MVF has relevant influence on the quantification of the motion corrected PET images in comparison to non-optimized MVFs. Therefore the optimized parameter set and non-optimized ones are evaluated.
For this evaluation static and dynamic data were acquired with an anthropomorphic thorax phantom. The reconstruction without motion correction shows a decrease of the max and mean values of about 47% and a tripled metabolic volume in comparison with ground truth (GT). The motion-corrected reconstruction with the optimized MVF shows a deviation of 1.5% for the max and mean values and of 21% for the volume compared to GT. In contrast to this, the reconstructions with the non-optimized MVFs show decreases of 38% - 58% for max and mean values and 2 - 4.5 times larger volumes. Thus, the reconstruction after optimizing the MVF shows a strong accordance with the GT values in contrast to the one with a non-optimized MVF. The validation with phantom data shows a clear influence of the parameters on the MVFs and further on the motion-corrected reconstructions. Therefore it is very important to generate MVFs that cover the existing physiological respiratory motion.

Keywords: Optical Flow, PET/CT, Motion Correction, Phantom study
Poster panel: 298

Poster Number:

Non-Invasive PET Head-Motion Correction via Optical 3D Pose Tracking  (#1570)

J. Goddard1, M. Mandelkern2

1 Innovative Vision Solutions, Knoxville, Tennessee, United States of America
2 VA Greater Los Angeles Health System, Los Angeles, California, United States of America


We have developed a method for motion correction in PET-CT brain scanning that relies on optical pose tracking of natural facial features and multiframe image reconstruction. No fixtures or markers are affixed to the subject. A compact stereo camera is mounted within the bore of a Siemens Biograph mCT PET-CT scanner. Images are acquired at 6–30 frames/sec. Scanner acquisition start and stop times are recorded to achieve synchronization between the scanner list file and image pose sequence. Face detection is used to locate head regions for 3D measurement. Pose tracking is performed to an initial reference frame. The pose data, in the form of a 4-dimensional rotation-translation matrix, are analyzed to identify quiescent intervals in which the pose x y z translations vary less than 1mm rms and the pose Euler angles vary less than 0.005 rms. Reconstruction is accomplished by dividing the scan into a sequence of sinograms with acceptable motion and incorporating the pose information iteratively in the reconstruction. Manikin studies with attached point sources demonstrate that repositioning is achieved with rms uncertainty < 0.5 mm. Human subject studies demonstrate repositioning to ~ 1.0 mm rms.

Keywords: Brain, Image Reconstruction, Motion compensation, Nuclear Imaging-PET, motion tracking
Poster panel: 301

Poster Number:

Impact of Time-of-Flight on Respiratory Motion Modelling using Non-Attenuation-Corrected PET (#1710)

A. C. Whitehead1, 3, E. C. Emond1, N. Efthimiou2, A. Akintonde3, 1, B. F. Hutton1, J. McClelland3, K. Thielemans1

1 University College London, Institute of Nuclear Medicine, London, United Kingdom
2 University of Hull, PET Research Centre, Hull, United Kingdom
3 University College London, Centre for Medical Imge Computing, London, United Kingdom


Respiratory motion reduces image quality in PET. Unless gated CT or MR data are available, motion correction relies on registration of the PET data. To avoid mis-registration due to attenuation mismatches, most existing methods rely on pair-wise registration of Non-Attenuation-Corrected (NAC) PET volumes. This is a challenging problem due to the low contrast and high noise of these volumes. This paper investigates the possibility of using motion models for respiratory motion correction in PET, and in particular whether incorporating Time-of-Flight (TOF) information increases the accuracy of the motion models derived from the NAC reconstructed images. XCAT phantom simulations are used for one bed position with a field of view including the base of the lungs and the diaphragm. A TOF resolution of 375ps is used. NAC images are reconstructed using OSEM and used as input for motion model estimation. Different motion models are compared using the original XCAT input volumes. The results indicate that TOF improves the accuracy of the motion model considerably.

Keywords: PET, Respiratory Motion Modelling, Non-Attenuation Corrected, Time of Flight
Poster panel: 304

Poster Number:

Assessment of Quantification Accuracy with ML Scatter Scaling for variable count statistics (#1822)

H. Bal1, V. Panin1, M. Conti1

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


Quantitative PETCT imaging requires the correction of scattered coincidences in the net trues data. Typically, this is done using single scatter simulation (SSS) based model to estimate the scatter shape followed by scaling to the net trues data via scatter tails which is referred to as the tail-fitted scatter scaling (TFSS). An alternate to TFSS is the maximum likelihood scatter scaling (MLSS) which employs an iterative update solution to compute the scaling factors along with an image update. In this work, we make further improvements to the MLSS method to improve quantification accuracy over a wide range of count statistics. Poisson ML based solution with nested loop for SSS scaling factors estimation was used to estimate scaling factors with MLSS. The number of nested loop iterations and the reconstruction iterations were incremented until the change in scaling parameters between successive iterations was minimal. Further, in order to reduce bias in the scaling parameters at low count statistics, image regularization was performed prior to executing MLSS algorithm. Assessment of quantification accuracy in the updated MLSS method was done using experimental phantom studies. These included a large uniform cylinder (37 cm diameter; 26 cm long) filled with F-18 of known concentration as well as a cardiac phantom with activity filled only in the cardiac insert. The list-mode PET data were rebinned into sinograms to obtain different count statistics and processed with MLSS and TFSS approach. Scatter fractions obtained with both approaches were compared in each case with respect to its high count dataset. Further, the PET activity concentration was compared with respect to the true activity concentration. MLSS with regularization was found to provide improved quantification accuracy for large degree of count statistics compared to TFSS approach in both phantom datasets. MLSS has the ability to provide good quantification in the case of low count PETCT imaging.

Keywords: PETCT, scatter, quantitative, low count
Poster panel: 307

Poster Number:

Dead-time correction method for block detector based ultra-long axial FOV PET scanner (#2051)

Y. Liu1, S. Tang1, J. Wang1, Y. Dong1

1 United Imaging Healthcare, CT&MI, Shanghai, China


The application of PET technology in preclinical studies and tumor diagnosis relies on accurate quantitative image reconstruction. For an ultra-long axial FOV PET scanner, a position-dependent dead-time correction is necessary due to different relative count rate load on the different parts of the detector system. Therefore, a component dead time correction model taking into account singles losses, coincidence losses, and the mismatch between the fixed LLD/ULD energy window and drifting energy spectrum is proposed to compensate the coincidence event losses in the scan process. In this method, a block-based non-paralyzable dead time model is used to compensate the single losses, a global paralyzable model as a function of module event rates is used to compensate the coincidence losses, we use the least squares fit to get the correction factor for each block because the detectors at ends of the scanner cannot get enough counts to ignore the background, and we also consider the variation of the correction factor for objects with different single-to-coincidence ratio. Additional correction related to the scatter fraction was performed to ensure the linearity of real coincidence events vs effective activity concentration. A series of decaying source experiments were finished to verify the validity of this method. Good response linearity is obtained for uniform water cylinder and NEC scatter phantoms with different lengths.

Keywords: quantitative imaging, total-body PET, dead-time correction, single-to-coincidence ratio
Poster panel: 310

Poster Number:

Maximum Likelihood Estimation of the Geometric Sensitivities in PET (#2119)

A. Rezaei1, T. Deller2, G. Schramm1, K. Wangerin2, F. Jansen2, K. Van Laere1, J. Nuyts1

1 KU Leuven, UZ, Medical Imaging Research Center, Leuven, Belgium
2 GE Healthcare, PET/MR engineering,, Waukesha, United States of America


In this work, we propose a maximum likelihood method to compute geometric and efficiency related correction from PET emission data.  In order to reduce the number of unknowns being estimated, a component and crystal based model is used for the geometric and efficiency based corrections, respectively. In a long acquisition of a uniform phantom scan, we show how possible inaccuracies in the projector or the scanner geometry can create observable artefacts in the reconstructions. Once the geometric and efficiency related corrections are re-estimated with our in-house projector the artifacts are eliminated.

Keywords: PET, Crystal Efficiencies, Geometric Sensitivities
Poster panel: 313

Poster Number:

Characterization of the Intel RealSense D415 Stereo Depth Camera for Motion-Corrected CT Imaging (#2177)

M. Dashtbani Moghari1, P. Noonan2, D. L. Henry3, R. Fulton3, 4, N. Young4, K. Moore4, A. Kyme1, 3

1 University of Sydney, Biomedical engineering, Sydney, Australia
2 King’s College London, School of Biomedical Engineering & Imaging Sciences, London, United Kingdom
3 University of Sydney, Health science, Sydney, Australia
4 Westmead hospital, Sydney, Australia


A combination of non-contrast CT (NCCT) and CT Perfusion (CTP) imaging is the most common regimen for evaluation of stroke patients. CTP-based image analysis is known to be compromised by patient head motion, however perfusion software packages typically only perform very basic frame-to-frame motion correction. Moreover, although intra-frame head motion is common, there are currently no techniques to perform intra-frame motion correction in CTP. Thus, the aim of this work is to investigate the feasibility of using the small form factor Intel RealSense D415 stereo depth camera to obtain accurate and rapidly sampled head pose estimates for intra-frame motion correction in CTP. First, we evaluated head movement quantitatively in a cohort of 72 acute stroke cases. Then we characterized the performance of the Intel D415 against ground-truth robotic motion and the clinically validated OptiTrack marker-based motion tracking system. The results showed that head motion during CTP imaging of acute stroke of patients is extremely common, with around 50% of patients moving > 5 mm and 1 deg and around 20% moving 10-100 mm and rotating 3-20 deg. The pose accuracy of the Intel for controlled robotic motion was approximately 5 mm and 2 deg. For translations and rotations, respectively. For human head motion using the OptiTrack as ground truth, the accuracy was approximately 4 mm (except for lateral motion) and 1.25 deg, respectively. Although poorer than what is needed clinically, there is a lot of potential to optimize performance and potentially achieve an accuracy consistently around 1 mm and 1 deg. Several improvements involving multiple cameras, improved region choice, and optimized tracking parameters are currently being investigated to achieve this.

Keywords: Motion tracking, Intel D415 Stereo depth camera, CT perfusion
Poster panel: 316

Poster Number:

Mass Preservation for Respiratory Motion Registration in both PET and CT (#2211)

E. C. Emond1, A. Bousse2, L. Brusaferri1, A. M. Groves1, B. F. Hutton1, K. Thielemans1

1 University College London, Institute of Nuclear Medicine, London, United Kingdom
2 Université de Bretagne Occidentale, LaTIM, INSERM, UMR 1101, Brest, France


Registration of medical images corresponding to different respiratory states is complicated, as respiration not only moves the lungs and nearby organs, but it also causes localised density and radiotracer activity concentration changes of up to 20% in the lung. Nevertheless, few registration methods incorporate mass-preserving constraints, which could lead to sub-optimal estimation of the deformation. This is especially important in the case of diffuse lung diseases for which the deformation field could provide biomarkers for pulmonary mechanical properties. This preliminary work is aimed at evaluating the impact of lung expansion in CT and PET image registration. We use patient data for which gated CT and PET acquisition data are available. We compare results of mass-preserving registration with two novel priors on the Jacobian determinant of the deformation field. Best results were observed for edge-preserving regularisation, where the mean errors in the lungs are decreased (-15.3% for CT and -30.6% for PET), compared to warped images with no mass preservation.

Keywords: Motion Compensation, Registration, Quantification
Poster panel: 319

Poster Number:

Compensation of Head Motion in Multi-Pinhole Brain SPECT Using a GPU-Based Iterative Reconstruction Algorithm (#2357)

N. Zeraatkar1, C. Lindsay1, B. Auer1, L. R. Furenlid2, P. H. Kuo2, M. A. King1

1 University of Massachusetts Medical School, Department of Radiology, Worcester, Massachusetts, United States of America
2 University of Arizona, Department of Radiology, Tuscan, Arizona, United States of America


Patient motion and its deteriorating effects in medical imaging is well known. Likewise, head rigid-body motion degrades the image quality in brain SPECT. We developed an algorithm to compensate the head motion in multi-pinhole SPECT systems within a statistical iterative image reconstruction algorithm. Previously, volunteer’s head motion was recorded by Vicon MX visual tracking system for 10 minutes while laying inside a SPECT/CT gantry. We then divided the motion into 120 intervals, each 5 seconds long. AdaptiSPECT-C, a multi-pinhole multi-detector stationary SPECT system, we are developing for dedicated brain imaging was used for this study. We generated an XCAT voxelized brain phantom emulating the activity distribution of Iodine-123 N-isopropyl-4-iodoamphetamine (IMP) for brain perfusion scan. To simulate the data acquisition with head motion, we used generic analytic simulation software we developed for multi-pinhole SPECT systems. The 6-degrees-of-freedom (6-DOF) motion was incorporated into the simulation software to realistically simulate the data acquisition with motion. Our previously developed graphics-processing-unit (GPU)-based iterative reconstruction software was augmented to incorporate motion compensation using 3D Gaussian interpolation. The rigid-body (i.e. 6-DOF) head motion was input to the reconstruction software through 120 motion intervals. For comparison, we reconstructed the motion corrupted SPECT data without motion compensation and a motion-free acquisition as ground truth. The results show that our proposed motion compensation method provides a significantly better SPECT reconstruction when compared to no motion compensation. The developed software can be applied for any scan duration with any number of motion intervals.

Keywords: AdaptiSPECT-C, brain perfusion SPECT, Gaussian interpolation, motion correction, motion compensation
Poster panel: 322

Poster Number:

Investigation of Optimal Respiratory Gating Methods for Continuous-Bed-Motion Whole-Body PET/CT (#2447)

Y. - J. Tsai1, Y. Lu1, J. Wu1, H. Liu1, P. Schleyer2, M. Casey2, C. Liu1

1 Yale University, Yale PET Center, New Haven, Connecticut, United States of America
2 Siemens, Siemens Medical Solutions, Knoxville, Tennessee, United States of America


Positron emission tomography (PET) scanners with continuous-bed-motion (CBM) feature that allows efficient whole-body dynamic scan and customized imaging protocol have been made commercially available in recent years. As the bed motion has to be reasonably slow in order to achieve sufficient count statistics, the blurring effect induced by respiratory motion should be considered and adequate gating methods should be applied. The aim of this study is to investigate and identify the reliable gating methods specific for CBM whole-body scans. Two methods, equal amplitude gating (EAG) and phase gating (PG), developed for single-bed or step-and-shoot multi-bed imaging protocol are evaluated using CBM data for patients with and without long-term respiratory baseline shift. Based on the results, we suggest the PG method is relatively reliable, while the EAG method introduces severe artifacts in the presence of baseline shift. Future work includes exploring the use of advanced motion compensated image reconstruction methods and performing quantitative evaluations on the gated images in terms of noise property and motion reduction.

Keywords: continuous-bed-motion, respiratory gating
Poster panel: 325

Poster Number:

The Initial Evaluation of An SRM-Based PET Normalization Method (#2538)

B. Li1, 2, B. Zhang1, L. Yang3, L. Fang1, C. Xu3, L. Xie4, Q. Xie1, 5, P. Xiao1, 5

1 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
2 Suzhou Raycan Technology Co., Ltd., Suzhou, China
3 Hubei RaySolution Digital Medical Imaging Technology Co., Ltd., Ezhou, China
4 Central Theater Command General Hospital of The Chinese People’s Liberation Army, Wuhan, China
5 Wuhan National Laboratory of Optoelectronics, Wuhan, China


The PET system response homogeneity is closely related to the manufacturing accuracy and the SRM modeling consistency. The physical modeling errors such as the crystal septa efficiency deviation and the mechanical error always lead to response heterogeneity. We have proposed an improved CBN algorithm for 3D PET normalization, which is named the SRM based component eliminated normalization (SCEN) method. The overall idea of this work is to eliminate the image artifact caused by system modeling error and reduce the interference effects between different components. We investigated the SCEN method based on the DigitMI 910 all-digital PET/CT scanner with phantom data preliminarily. The results show that the background variability is reduced by 21.0%~24.8% using the SCEN method.

Keywords: PET normalization, all-digital PET, data process, image reconstruction
Poster panel: 328

Poster Number:

Barreloid Deformation Correction in Planar Imaging (#2595)

D. Zarketan1, 2, M. - E. Tomazinaki1, E. Stiliaris1, 3

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


Unbalanced amount of the accumulated charge on the anodic grid of a Photomultiplier Tube combined with Center of Gravity position reconstruction algorithms produces systematically shifts towards the center of the image, reducing the effective field of view and causing deformation effects at the edges of the image. The main purpose of this study is to detect the origin of this kind of image deformation, mainly known as barreloid distortion, and in addition to propose appropriate algorithmic approaches for the correction of these effects on planar imaging. Various analyses with simulated and real data will be presented and the success of the method will be discussed with different phantom geometries measured with a small γ-Camera system.

Keywords: Center of Gravity Algorithm, Position Reconstruction, Planar Imaging, γ-Camera System
Poster panel: 331

Poster Number:

Data-driven Estimation of Crystal Efficiencies using Single Events (#2613)

T. Feng1, A. Selfridge2, L. He3, E. Leung2, Y. Liu3, B. Spencer2, J. Schmall1, J. Qi2, S. R. Cherry2, R. D. Badawi2

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


Normalization correction is a pre-requisite for accurate reconstruction of PET images. Uniform phantoms are often used to estimate normalization factors. While this approach can provide reliable results, the additional phantom-based scans reduce overall patient throughput.
The main component in the normalization factors that may change over time is the crystal efficiencies. The crystal efficiencies affect both coincidence events and single events. Single events are used to provide additional data for estimating crystal efficiencies. A maximum-likelihood-based approach has been derived and used for the estimation of crystal efficiencies, which indicates that the crystal efficiency can be derived by dividing the measured single events with the estimated single events from activity distribution and attenuation map. An alternate update approach has been developed, where the activity distributions and the crystal efficiencies are jointly estimated. 2D simulations with different noise levels were employed to validate the method. Phantom scan using the partially developed EXPLORER system (two detector units) were also carried out. Fast Monte Carlo simulations were applied for acquiring the estimated single events from activity distributions.
Noise-free simulations suggested that the proposed method can quantitatively recover the unknown crystal efficiencies. Noisy simulations suggested that high accuracy can still be achieved (~1% error) with noisy data. The study using the physical scan indicated that it is practical to acquire the measured and estimated singles, and the crystal efficiency map can be acquired using our method.
We have proposed a method that can estimate crystal efficiencies using single events and routine clinical data. More studies using acquired data from the scanner will be conducted to further validate the method. This method makes it possible to reduce the number of phantom-based scans for quality control purposes.

Keywords: Crystal efficiency, single events, PET, Normalization
Poster panel: 334

Poster Number:

Study of Repeatability of a Novel PET Flow Phantom (#1453)

R. Siekkinen1, 2, J. Teuho1, A. K. Kirjavainen3, K. Koskensalo1, 2, A. Saraste1, 4, M. Teräs2, 5

1 Turku University Hospital and University of Turku, Turku PET Centre, Turku, Finland
2 Turku University Hospital, Department of Medical Physics, Division of Medical Imaging, Turku, Finland
3 University of Turku, Radiopharmaceutical Chemistry Laboratory, Turku PET Centre, Turku, Finland
4 Turku University Hospital, Heart Center, Turku, Finland
5 University of Turku, Institute of Biomedicine, Turku, Finland


Dynamic PET imaging is a valuable method in myocardium perfusion imaging. The quantification of blood flow in the myocardium can be performed based on the tracer distribution. To develop myocardium perfusion imaging further, a reference test object modelling the flow in the myocardium is needed. A newly developed PET flow phantom provides a validated platform for blood flow quantitation with a known reference flow. However, since the reference test object is considered as the ground truth in flow quantitation, the repeatability of the flow phantom derived flow values should be confirmed.
In this study, we minimized the external sources that might have an influence in the image-derived flow values and studied the repeatability of the phantom derived flow values. The aim of the study was to confirm that no significant variability existed between the image-derived and reference flow values between subsequent repeats. This study was conducted with a flow phantom in test-retest protocol with one test scan and four re-test scans. The imaging conditions were stable during the measurements. The PET images were analyzed with a phantom-vendor provided software to derive the phantom time-activity curves and flow values.
The image-derived flow values varied with STD of 0.89 mL/min and 0.86 mL/min between subsequent repeats. The error between the image-derived flow values and the reference flow values were less than 5.5 % within all subsequent repeats. Sources affecting the variations in the flow values were considered to arise from the variations in the administered dose rates and fluctuations in the water flow inside the phantom. Overall, the phantom showed high repeatability in flow values.
In conclusion, the newly developed PET perfusion flow phantom shows high repeatability in the image-derived flow values and is a reliable platform for quality control and validation studies in myocardial perfusion imaging in PET.

Keywords: perfusion, quantification, repeatability, PETCT, phantom
Poster panel: 337

Poster Number:

Blind CT Image Quality Assessment via Deep Learning Framework (#2026)

Q. Gao1, 2, S. Li1, 2, M. Zhu1, 2, D. Li1, 2, D. Zeng3, 1, Z. Bian1, 2, Q. Lyu4, 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 South China University of Technology, College of Automation Science and Engineering, Guangzhou, China
4 Southern Medical University, Zhujiang Hospital, Guangzhou, China

This work was supported in part by the National Natural Science Foundation of China under Grants 61701217, 81701690, 61571214 and U1708261, the Science and Technology Program of Guangzhou, China under Grant 201705030009, the Science and Technology Program of Guangdong, China under Grant 2015B020233008.


Computed tomography (CT) images will be severely damaged from low-mAs acquisition conditions. Seriously degraded CT images may lead to diagnostic bias in clinics. It is vital to assess CT image quality before diagnosis. However, lack of high quality reference CT images in clinical practice makes the full-reference (FR)- and reduced-reference (RR)- image quality assessment (IQA) models difficult for CT image quality assessment. In addition, convolutional neural network (CNN) has been proven effective for natural image quality assessment. Meanwhile these CNN-based methods require a large amount of manually labeled data for network training to get satisfactory results. It should be noted that the pre-collected dataset is limited and manual labeling is time-consuming and labor-intensive. In this work, we develop a new no-reference (NR)-IQA strategy in a deep learning framework to assess CT image quality effectively. Specifically, at first, we use the FR-IQA metric, i.e., PSNR and SSIM, to comprehensively assess CT image quality objectively which is served as label for NR-IOA network. The PSNR and SSIM measures the noise level and structural distortion in the CT images which are the two most vital factors affecting diagnosis. Then, we employ the label data to train a convolutional neural network, making it capable of providing quality measurements for each input image without reference image. The experimental results with Mayo dataset demonstrate that the present NR-IQA network can accurately predict CT image quality.

Keywords: image quality assessment, Computed tomography, no-reference, full-reference, deep learning
Poster panel: 340

Poster Number:

Characterisation of PET/MR scanners for brain imaging in Dementias Platform UK clinical trials (#2696)

P. J. Markiewicz1, 5, J. C. Matthews2, A. Barnes3, J. Dickson3, D. Thomas1, J. Davies4, W. Hallett4, G. Krokos5, J. Mackewn5, P. Marsden5, E. De Vita5, J. Sterling5, G. Delso6, G. Macnaught7, T. Clark7, C. Wimberley7, G. Thompson7, J. M. Anton-Rodriguez2, I. Vamvakas2, A. Watkins2, R. Maxwell8, E. Howell8, R. Manavaki9, T. D. Fryer9, V. Lupson9, M. Mada9, V. Rhodes-Bradford2, J. Wardlaw7, J. - P. Taylor8, J. O'Brien9, A. Hammers5, S. Ourselin5, N. C. Fox1, K. G. Herholz2, F. Barkhof1, 10

1 University College London, London, United Kingdom
2 University of Manchester, Manchester, United Kingdom
3 University College London Hospital Trust, London, United Kingdom
4 Invicro, London, United Kingdom
5 King’s College London, London, United Kingdom
6 GE Healthcare, Barcelona, Spain
7 University of Edinburgh, Edinburgh, United Kingdom
8 Newcastle University, Newcastle, United Kingdom
9 University of Cambridge, Cambridge, United Kingdom
10 VU Amsterdam, Amsterdam, Netherlands


The purpose of this work is to characterise and compare image quality for quantitative brain PET imaging using the Dementias Platform UK (DPUK) network of seven PET/MR scanners, and which consists of three Siemens Biograph mMR and four GE Signa scanners. A set of phantom scans were performed on all scanners with three main aims: (1) To provide baseline performance measurements from which future qualification standards can be set for the use of PET/MR scanners in clinical trials; (2) to provide an understanding of any differences observed with the clinical amyloid brain test-retest data, which follows this phantom study; and (3) to provide the assessment of the utility and logistics of performing phantom measurements for PET/MR. Three physical phantoms were used: (1) a long 10L uniform cylindrical phantom to assess the PET performance across the axial field of view (FOV≥25 cm), including activity outside the FOV; (2) a 5L bottle phantom to assess the PET performance with the use of head and neck coils while running simultaneously high SAR/gradient duty cycle MR sequences; (3) a 6.4L ACR-approved Jaszczak PET phantom to assess image resolution and contrast. The filling of all the phantoms and data acquisition was performed at all sites under the supervision of a single operator (PJM), to ensure scanning harmonization. The attenuation correction was performed using CT-based attenuation maps. The preliminary results indicate good uniformity across the transaxial and axial FOV, with some mild but noticeable axial streaks, possibly due to the inaccuracies in attenuation correction for the patient bed in both scanners. The effects of intense MR sequences are detectable but within the test/retest variability of PET. The activity outside the FOV had a detectable effect on the contrast and background activity. The attenuation correction for the hardware (bed and head coils) was overall adequate for all scanners.

Keywords: PET MR, phantom, analysis, multicentre, clinical trials
Poster panel: 343

Poster Number:

A novel convolutional neural network for predicting full dose from low dose PET scans (#2756)

A. Sanaat1, H. Zaidi1, 2

1 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Genève, Switzerland
2 University of Geneva, Geneva Neuroscience Centre, Geneva, Genève, Switzerland


The use of radiolabelled tracers in PET imaging raises concerns owing to potential risks from radiation exposure. Therefore, to reduce this potential risk in diagnostic PET imaging, efforts have been made to decrease the amount of radiotracer administered to the patient. However, by decreasing the injected dose, the signal-to-noise Ratio (SNR) decreases and image quality deteriorates, thus adversely impacting clinical diagnosis. Previously proposed techniques are complicated and slow, yet still do yield satisfactory results at significantly low dose. In this work, we propose a deep learning algorithm to reconstruct standard-dose from low-dose PET images using a fully convolutional encoder-decoder deep neural network model. The goal is to train a model to learn to reconstruct from images with only 10% counts to produce images corresponding to 100% of the dose. Brain PET/CT images of 15 patients acquired on the Siemens Biograph mCT with a standard dose of 18F-FDG (200 MBq). Images were acquired for about 20 min. The sinograms of each scan were used to produce a low-dose sinogram by randomly selecting only 10% of the counts. To avoid overfitting, data augmentation was used. The adopted CNN is based on an encoder-decoder structure, each stage consisting of a convolution with 2×2 kernels, batch normalization, and rectified linear unit (ReLU). The downsampling and upsampling between stages are done by 2×2 max pooling. Detailed quantitative and qualitative comparison demonstrated the proposed method can generate artefact-free diagnostic quality images that preserve internal structures without noise amplification. The structural similarity index (SSIM) and peak signal to noise ratio (PSNR) were used as quantitative metrics for assessment. For instance, the PSNR and SSIM in selected slices were 40 and 0.97, respectively. The proposed algorithm operates in the projection space and is capable of producing diagnostic quality images with 10% of the standard injected dose.

Keywords: convolutional neural network, Low Dose, Sinogram, reducing injected dose, PET
Poster panel: 346

Poster Number:

Pseudo-proton dose generation using prompt gamma imaging with deep learning (#1068)

C. - C. Liu2, H. - M. Huang1

1 National Taiwan University, Institute of Medical Device and Imaging, Taipei City, Taiwan
2 University of California, Davis, Department of Biomedical Engineering, Davis, California, United States of America


In proton therapy, dose verification is important. However, measuring dose distribution is difficult. In this work, we study the feasibility of using deep learning to generate proton dose distribution from proton-induced prompt gamma (PG) imaging. We simulated 20 brain phantoms irradiated with a 100-MeV proton pencil beam. For each phantom, we simulated 200 PG images and proton dose distributions (10 entrance locations 20 beam sizes). A convolutional neural network based on the U-Net architecture was trained to generate pseudo-proton dose distributions from PG images. Our simulation results show that the pseudo-proton dose distribution derived from PG image agrees well with the simulated true one. In addition, the pseudo-proton dose distribution can predict the Bragg peak position with a mean accuracy of less than 1.0 mm. Next, we will investigate how many PG images are sufficient for training a reliable model. Moreover, we will investigate whether the learned model can work well with unseen simulation conditions including different beam sizes, proton energies and phantoms.

Keywords: proton dose, deep learning, prompt gamma imaging
Poster panel: 349

Poster Number:

Laboratory and test-beam results with MACACO II (#1241)

A. Ros Garcia1, L. Barrientos1, J. Barrio1, J. Bernabéu1, M. Borja-Lloret1, P. Dendooven2, F. J. García López3, M. C. Jimenez-Ramos3, C. Lacasta1, R. Marco1, E. Muñoz1, J. F. Oliver1, I. Ozoemelan2, J. Roser1, C. Solaz1, R. Viegas1, G. Llosa1

1 Instituto de Fisica Corpuscular (CSIC-UV), Valencia, Spain
2 University of Groningen, KVI - Center for Advanced Radiation Technology, Groningen, Netherlands
3 Centro Nacional de Aceleradores, Universidad de Sevilla, Sevilla, Spain


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 LaBr 3 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. In addition, the image reconstruction code has been improved with an accurate model of the sensitivity matrix. Furthermore, a new spectral reconstruction algorithm for Compton cameras with two planes has been developed showing promising results. The system has been fully characterised in the laboratory at controlled temperature and also 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. The results show that for measurements taken in different positions of the PMMA target the reconstructed Bragg peak position is related to the PMMA position.

Keywords: hadrotherapy, compton imaging, SiPM, reconstruction, test-beam
Poster panel: 352

Poster Number:

Estimation of Cerenkov light distribution in the luminescence image of water for carbon-ion therapy dosimetry (#1380)

T. Yabe1, 2, Y. Hirano1, M. Komori1, T. Akagi3, S. Yamamoto1

1 Nagoya University Graduate School of Medicine, Radiological and Medical Laboratory Sciences, Nagoya, Japan
2 Nagoya University Hospital, Department of Medical Technology, Nagoya, Japan
3 Hyogo Ion Beam Medical Center, Department of Radiology, Tatsuno, Japan


Recently, we successfully imaged the luminescence of water using a cooled charge-coupled device camera during irradiations of therapeutic proton and carbon-ions. The depth profiles obtained from the luminescence images showed similar distributions to dose. However, for the measured image of 245 MeV/n carbon-ions, the depth profiles showed high intensity in the shallow area of the image mainly due to the Cerenkov light because energy of 245 MeV/n carbon-ions is higher than the Cerenkov light threshold for the produced secondary electrons to emit Cerenkov light. Also, the beam width of the luminescence image was larger than that measured by the ionization chamber and showed an offset luminescence at the bottom of the lateral profiles. We hypothesized that this phenomenon was caused by the Cerenkov light produced by the interaction of prompt gamma photons and other particles with water. In this study, we estimated the distribution of Cerenkov light produced in water using Monte Carlo simulation (Geant4) during irradiations of 245 MeV/n carbon-ions. Subsequently, we corrected the depth and lateral profiles obtained from the measured image by subtracting the simulated distribution of Cerenkov light . The corrected depth and lateral profiles of the luminescence image were compared with the dose distributions. With the corrections, the profiles of the luminescence image had almost identical distributions to the dose distributions for carbon-ions. In the shallow area, the depth profile with correction was almost the same curve with the difference from dose distribution within 6%, which was within 133% without correction. The lateral profiles with correction also had almost the same widths with the difference from dose distribution within 2%, which was within 17% without correction. These results indicate that the luminescence imaging of water has a potential to be used for the dose distributions measurement of carbon-ion therapy.

Keywords: luminescence, imaging, Cerenkov-light, Monte Carlo simulation, carbon-ion therapy
Poster panel: 355

Poster Number:

Proton Beam Range Verification with Secondary Radiation from Titanium Implants (#1468)

C. M. Bäcker1, 2, C. Bäumer2, A. Bley1, P. Fragoso Costa3, 4, M. Gerhardt1, K. Herrmann3, 4, S. Kauer1, 2, K. Kröninger1, C. Nitsch1, H. M. Siregar1, 2, B. Timmermann2, 5, A. Yazgan1, 2

1 TU Dortmund University, Experimentelle Physik IV, Dortmund, North Rhine-Westphalia, Germany
2 West German Proton Therapy Centre Essen and West German Cancer Center, Essen, North Rhine-Westphalia, Germany
3 University Hospital of Essen, Clinic for Nuclear Medicine, Essen, North Rhine-Westphalia, Germany
4 West German Cancer Center, Essen, North Rhine-Westphalia, Germany
5 University Hospital of Essen, Clinic for Particle Therapy, Essen, North Rhine-Westphalia, Germany


Several methods of proton beam range verification have been evaluated during the last decades. Since so-called in-room solutions require the space in the treatment room and off-line imaging methods, e.g. positron emission tomography (PET), is limited by the wash out of biological molecules, another method is evaluated during this experimental study on the potential for PET imaging. A possible solution for the biological wash out is that, four out of five brain tumor patients at the West German Proton Therapy Centre Essen (WPE) have titanium clips from surgical resection. This requires a precise knowledge of the produced radionuclides from titanium after the irradiation with protons and the corresponding activation cross sections. For the determination of the activation cross section a pure, thin titanium sheet is irradiated. A titanium alloy, which is called grade 5 and used for implants and clips, is also irradiated and compared with the pure one. Furthermore, a first simplified titanium based phantom is irradiated and afterwards a PET scan is taken to determine the activity with the sample. The first studies show the potential of titanium for proton beam range verification.

Keywords: Biomedical applications of radiation, Gamma-ray detection, Nuclear Imaging, Radiation monitoring
Poster panel: 358

Poster Number:

A comparison of direct reconstruction algorithms in proton CT (#1509)

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

1 Univ Lyon, INSA-Lyon, Université Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR 5220, U1206, Lyon, France


Several analytic algorithms have been proposed to incorporate the non-linear path of protons in the reconstruction of a proton CT (pCT) image. This paper presents a comparison between four direct algorithms used in pCT, in terms of spatial resolution and relative stopping power (RSP) accuracy. We have simulated a pCT set up which registers protons individually using Gate, a Monte Carlo simulation tool, with a 200 MeV proton source and two position, direction and energy detectors upstream and downstream from the object. A Catphan 528 phantom and a spiral phantom were imaged to take into account the spatial dependency of the spatial resolution. Each proton’s trajectory was estimated using a most likely path (MLP) formalism. The spatial resolution was evaluated using the frequency corresponding to an MTF value of 10%, and the RSP accuracy as the mean value in a homogeneous region. Our results show that methods performing the backprojection before the filtering offer a better spatial resolution (up to +36%) since each proton is directly binned in the image grid according to its MLP. However, this improvement is minor (+2%) at the center of the object, where the intrinsic uncertainty on the MLP estimate is dominant. Regarding the RSP accuracy, all algorithms but one show equivalent results.

Keywords: proton computed tomography, proton CT
Poster panel: 361

Poster Number:

Optimization of Coincidence Time Window on Compton Imaging of Astatine-211 for Targeted α-Particle Radiotherapy (#1628)

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

1 National Institutes for Quantum and Radiological Science and Technology (QST), Takasaki Advanced Radiation Research Institute, Quantum Beam Science Research Directorate, Takasaki, Japan
2 Tohoku University, Cyclotron and Radioisotope Center, 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 single-photon emission computed tomography or gamma cameras, it is important to image 211At directly. The 211At and its daughter radionuclide 211Po emit gamma rays (570 keV, 687 keV, and 898 keV) at the total intensity of 0.9%. Recently, we have proposed to image 211At with the gamma rays using a Compton camera and demonstrated the imaging capability of the camera in the experiments of a point-like 211At source with a relatively wide coincidence time window. Since chance coincidence events by polonium K-shell x rays were dominant and seemed to cause saturation of counts in the experiments, optimization of the coincidence time window is important to reduce the chance coincidence events. In this study, we optimized the coincidence time window and evaluated the performance of the camera. As a result, the sensitivity and images were improved. Future research plans include the evaluation of image quality.

Keywords: Astatine-211, Compton camera, Targeted radiotherapy
Poster panel: 364

Poster Number:

A new proton irradiation facility at Chang Gung Memorial Hospital in Taiwan (#2020)

C. - Y. Pan1, Y. - C. Tsai2, C. - H. Hsing1, C. - H. Wang3, H. Niu4, T. - Y. Hsiao4, T. - Y. Chen4, C. - H. Chen4, T. - C. Chao1, 5

1 Chang Gung University / Chang Gung Memorial Hospital, Medical Physics Research Center, Institute for Radiological Research, Linkou, Taoyuan, Taiwan
2 Chang Gung University / Chang Gung Memorial Hospital, Particle Physics and Beam Delivery Core Laboratory, Institute for Radiological Research, Linkou, Taoyuan, Taiwan
3 National United University, Department of electro-optical engineering, Miaoli, Taiwan
4 National Tsing Hua University, Accelerator Lab. Nuclear Science and Technology Development Center, HsinChu, Taiwan
5 Chang Gung University, Department of Medical Imaging and Radiological Sciences, Linkou, Taoyuan, Taiwan


The construction of a new facility dedicated to radiobiology and physics researches has been completed at the Chang Gung Memorial Hospital (CGMH) in Taiwan. There are three beam lines installed for broad and small field proton irradiation and high energy neutron generation. We present the status of the new proton irradiation facility with the proton beams energy ranges between 70 MeV and 230 MeV. A double scattering system was designed to generate an irradiation field that is large enough to ensure the delivery of dose greater than 90% uniformity over an area of several squared centimeters. The apparatus includes a wheel type range modulator for spread-out Bragg peak (SOBP), and a strip 2D ionization chamber and a multilayer ionization chamber for beam size and dose profiles in XY plane and Z direction, respectively. The primarily beam profile study results are very promising and can serve the purpose of proton or neutron beams radiations related research. This facility will soon be open world wildly and the user applications will be welcomed.

Keywords: proton therapy, proton irradiation
Poster panel: 367

Poster Number:

Determination of the photon interaction position in a monolithic scintillator via a Convolutional Neural Network based algorithm (#2151)

M. Kawula1, T. M. Binder1, 2, S. Liprandi1, K. Parodi1, P. Thirolf1

1 Ludwig-Maximilians University, Medical Physics, Garching bei München, Germany
2 KETEK GmbH, München, Germany


In this work we investigated an algorithm based on the Convolutional Neural Network (CNN) approach to determine interaction positions of  γ-quanta in monolithic scintillators, used as absorber component of a Compton Camera system under development for ion beam range  verification via prompt gamma imaging. Comparison is provided with the performance of a ‘conventional’ k-Nearest-Neighbour (k-NN) algorithm. Monolithic crystals are considered an interesting alternative to the pixelated arrays, able to provide very good timing and energy resolution as well as high sensitivity. In our work we were considering such a crystal blocks of LaBr3:Ce and CeBr3 each with dimensions of 50 mm × 50 mm × 30 mm. We tested the spatial resolution for three photon energies of 662 keV (137Cs), 1.17 and 1.33 MeV (60Co) via 2D detector scans with tightly collimated photon sources. The best results were obtained for the highest energy, reaching 1.2(1) mm and 1.3(1) mm (FWHM) for CeBr3 and LaBr3:Ce respectively, limited only by the collimator diameter of 1mm. The CNN algorithm has minimal memory requirements and allows reconstructing up to 104 events per second by using only one CPU, beneficial on the route towards a future in-vivo clinical applicability of the Compton Camera for ion beam range verification.

Keywords: Beam range monitoring, Compton Camera, machine learning, monolithic scintillator, neural network
Poster panel: 370

Poster Number:

Investigating the Performance of a Novel Hybrid Gamma-Imaging Technique Toward Future 3D Reconstruction of Proton Beam Range (#2385)

M. J. Safari1, A. Zoglauer2, G. Lovatti1, V. Anagnostatou1, M. Nitta1, H. Tashima3, T. Yamaya3, P. Thirolf1, G. Dedes1, K. Parodi1

1 Ludwig Maximilians Universität München, Department for Medical Physics, Munich, Bavaria, Germany
2 University of California, Space Sciences Laboratory, Berkeley, California, United States of America
3 National Institute of Radiological Sciences, National Institutes for Quantum and Radiological Science and Technology, Chiba, Japan


γ-PET imaging has been proposed as a novel imaging technique for 3D in-vivo ion beam range verification in particle therapy. By using the coincident detection of annihilation and prompt gamma Compton events, produced by target and projectile nucleus fragmentation, the radiation source position could be efficiently localized on the intersection region between the line of response (LOR) and the surface of the Compton conefor the third prompt photon. The goal of this work was to study the performance of Positron-Emission-Tomography (PET), Compton imaging, and combined γ-PET for reconstruction of a point β+(γ) emitter source (22Na). In this work, the Medium-Energy Gamma-ray Astronomy library (MEGAlib) simulation toolkit was used as a dedicated simulation and image reconstruction framework for PET, Compton, and γ-PET imaging. The scanner geometry used in this study resembled the Whole Gamma Imaging (WGI) scanner from NIRS-QST. A spherical PMMA phantomwith a 5 cm radius and a 22Na point source were used and positioned in the centre of the simulationgeometry.The angular resolution for the Compton scattering events and the PET detectors’ spatial resolution were evaluated individually. The spatial resolution of the reconstructed PET, Compton, and γ-PET events were obtained and compared. This study supports the promise of the proposed approach and motivates further studies to address the imaging performances and system optimization in more realistic in-beam scenarios.

Keywords: in-beam γ-PET, triple-γ imaging, ion range verification, MEGAlib
Poster panel: 373

Poster Number:

Prompt gamma-ray imaging for real-time in vivo range/dose verification in proton and carbon ion therapy (#2441)

M. Xiao1, S. Paschalis1, P. Joshi1, T. Price2

1 University of York, Department of physics, York, United Kingdom
2 University of Birmingham , School of physics and Astronomy, Birmingham , United Kingdom


In vivo range verification is desirable to understand the range uncertainties, minimizing beam delivery errors during hadron therapy. The aim of this project is to develop a novel prompt gamma-ray imaging (PGI) prototype detector for the absolute and relative range verification of hadron therapy, which can be used in clinical tests. The detection system consists of arrays of scintillator LFS (Luteium Fine Silicate) crystals coupled to silicon photomultiplier (SiPM) arrays. A passive tungsten multi-slit collimator has been constructed in-house and used for the PG profile along the depth of serval phantoms. In the meantime, the proton dosimetry is estimated using a microprobe that is inserted into the phantom. The correlated measurements from both detectors will be used for the determination of the relationship between “Bragg peak” (BP) and PG peak. We will present preliminary results from in-beam measurements using the MC40 cyclotron at Birmingham with a proton beam at 38 MeV and the KVI CART facility with proton and Carbon beams with energies of around 150MeV and 90 MeV/nucleon, respectively. Time-of-flight (ToF) resolution, energy response and gamma-ray profiles will be shown for an energy window corresponding to 3 – 7MeV gamma rays. The results are also compared to a Monte Carlo (MC) Geant4 simulation model that was developed within this project.

Keywords: prompt gamma imaging, Bragg peak, range verification.
Poster panel: 376

Poster Number:

Studies of J-PET detector to monitor range uncertainty in proton therapy (#2637)

A. Rucinski1, J. Baran1, M. Dadgar2, J. Gajewski1, M. Pawlik-Niedźwiecka1, 2, P. Moskal2

1 Institute of Nuclear Physics PAN, Proton Radiotherapy Group, Krakow, Poland
2 Jagiellonian University, Faculty of Physics, Astronomy and Applied Computer Science, Krakow, Poland


A novel detector techniques are needed to tackle the problem of range uncertainty to eventually increase the number of patients potentially benefiting from particle therapy. In Krakow, new, affordable, modular, lightweight, portable and reconfigurable technology of plastic scintillator based positron emission tomography detector (J-PET) was developed. A single J-PET module is constructed out of thirteen 50cm long scintillator strips. Radiation produces light pulses in a strip that are propagated to the strip edges and are converted to electrical signals with silicon photomultipliers read-out by fast on-board front-end electronics.  This design potentially enables efficient detection of beta-plus emitting isotopes induced by a therapeutic proton beam in a patient. We investigate the feasibility of cylindric, dual head, single, multi-layer J-PET configurations to design detector suitable for in-room proton therapy range monitoring. We performed Monte Carlo simulations to evaluate the detector signal per primary proton as a function of detector acceptance and efficiency. Currently, we are preparing experimental activities to validate Monte Carlo simulations using a therapeutic proton beam available in Krakow proton beam facility. In parallel, we perform Monte Carlo simulations to estimate correction factors as well as perform three-dimensional list-mode reconstruction of PET image using QETIR and CASTOR software packages. Our studies aim at a reliable estimation of sensitivity and precision of the detector to monitor range uncertainties in proton therapy. The current status and results of project activities will be presented. 

Keywords: positron emission tomography, proton therapy, radiation detectors, radiation imaging
Poster panel: 379

Poster Number:

Deep Learning for MRI-based CT Synthesis: a comparison of MRI Sequences and Neural Network Architectures (#1278)

A. Larroza1, L. Moliner1, J. M. Álvarez-Gómez1, S. Oliver1, H. Espinós-Morató1, M. Vergara-Díaz1, M. J. Rodríguez-Álvarez1

1 Universitat Politècnica de València (UPV)-Consejo Superior de Investigaciones Científicas (CSIC), Instituto de Instrumentación para Imagen Molecular (I3M), Valencia, Spain


Synthetic computed tomography (CT) images derived from magnetic resonance images (MRI) are of interest for radiotherapy planning and positron emission tomography (PET) attenuation correction. In recent years, deep learning implementations have demonstrated improvement over atlas-based and segmentation-based methods. Nevertheless, several open questions remain to be addressed, such as which is the best of MRI sequence and neural network architecture. In this work, we compared the performance of different combinations of two common MRI sequences (T1- and T2-weighted), and three state-of-the-art neural networks designed for medical image processing (Vnet, HighRes3dNet and ScaleNet). The experiments were conducted on brain datasets from a public database. Our results suggest that T1-weighted perform better than T2-weighted images, but the results further improve when combining both sequences. The lowest mean average error over the entire head (MAE = 101.76 ± 10.4 HU) was achieved combining T1 and T2 scans with HighRes3dNet. All tested deep learning models achieved significantly lower MAE (p < 0.01) than a well-known atlas-based method.

Keywords: Deep Learning, Magnetic Resonance Imaging, Positron Emission Tomography, Synthetic Computed Tomography
Poster panel: 382

Poster Number:

Generative adversarial network for denoising in dual gated myocardial perfusion SPECT using a population of phantoms and clinical data (#1606)

J. Sun1, Q. Zhang1, D. Zhang1, P. H. Pretorius2, M. A. King2, G. S. Mok1, 3

1 University of Macau, Department of Electrical and Computer, Taipa, China, Macao Special Administrative Region
2 University of Massachusetts Medical School, Department of Radiology, Worcester, Massachusetts, United States of America
3 University of Macau, Faculty of Health Sciences, Taipa, China, Macao Special Administrative Region


Previously we proposed to use a generative adversarial network (GAN) in denoising dual respiratory-cardiac gating (DG) images for myocardial perfusion SPECT. In this study we further compared the use of various training datasets and application on clinical data. Five 4D Extended Cardiac Torso phantoms with cardiac motion, different anatomies, respiratory characteristics and activity uptakes were used in the simulation, modeling 6 respiratory and 8 cardiac gates, i.e., a total of 48 DGs. One hundred and twenty noisy LEHR projections were generated analytically and then reconstructed by the OS-EM algorithm with 6 subsets and 5 iterations. A clinical dataset for a patient who underwent SPECT/CT 1 hr post injection of 1332 MBq Tc-99m sestamibi was re-binned into 7 respiratory and 8 cardiac gates, and then reconstructed by the ML-EM algorithm with 24 iterations. The GAN was implemented using Torch. Using patients’ own data, eighteen DG images were paired with the ungated or the corresponding cardiac gate for training. Using cardiac gate data as the training dataset, we also evaluated the use of other patients’ datasets for training by increasing the patient database from 1 to 4. The noise level measured as the normalized standard deviation (NSD) on a 2D uniform region of liver and the FWHM on the image profile drawn across the left ventricle wall were compared. In simulations, the NSD/FWHM (cm) for training using ungated and cardiac gate data were 0.050/1.627 and 0.110/1.489. They were 0.157/3.132, 0.152/2.646, 0.108/1.652 and 0.082/1.375 for training using one to four patients’ datasets respectively. The NSD/FWHM for 1 DG data before GAN were 0.275/1.373. The clinical data showed that the use of GAN can lower the noise (NSD 0.246 vs 0.110) with minimal degradation of resolution (FWHM 1.682 vs 1.912). The use of cardiac gate and patient’s own data for training provide superior denoising results. More phantom and patient data are warranted to confirm our findings.

Keywords: SPECT, Dual gating, Denoising, Generative adversarial network
Poster panel: 385

Poster Number:

A deep learning approach to fully-automated 3D segmentation of brain magnetic resonance images (#1825)

K. H. - Y. Leung1, 2, J. M. Coughlin2, S. P. Rowe2, M. G. Pomper1, 2, Y. Du2

1 Johns Hopkins University School of Medicine, Department of Biomedical Engineering, Baltimore, Maryland, United States of America
2 Johns Hopkins University School of Medicine, The Russell H. Morgan Department of Radiology, Baltimore, Maryland, United States of America


Quantitative analysis of brain SPECT and PET images are routinely performed for neurological disorders. Reliable definition of volumes-of-interest from MRI images is a critical step in such quantitative analysis. While manual segmentation is considered the gold standard, it is tedious, time consuming, and suffers from inter- and intra-operator variability. Unsupervised automatic segmentation methods are typically slow and sensitive to noise, artifacts, and contrast variation in the images. We propose a deep learning approach to fully-automated 3D segmentation of the brain. We investigated segmentation of the right and left regions of the caudate and putamen. A total of 192 brain MR images were segmented by FreeSurfer. The FreeSurfer segmentations were used as ground truth to train the deep learning approach. The data were randomly partitioned into training, validation and test sets using a training/validation/test split of 60%/20%/20%. The 3D U-net were optimized to the validation set. The network was trained by minimizing a cross-entropy loss function with Adam, a first-order stochastic gradient-based optimization algorithm. The optimized 3D U-net was trained on 154 images and then evaluated on 38 independent images. The proposed deep learning method yielded a Dice similarity coefficient (DSC) of 0.87±0.03 and a Jaccard similarity coefficient (JSC) of 0.77±0.05 for all striatal regions. DSC and JSC are metrics that describe overlap where higher values indicate more accurate segmentation. The true positive fraction and true negative fraction were calculated as 0.92±0.03 and 1.00±0.00, respectively. Those results demonstrate that the proposed method is accurate. The proposed method also provided fast segmentation taking 0.77±0.04 seconds per image. Visually, the delineated brain regions by the proposed method had high overlap with the ground truth. The proposed deep learning approach showed significant promise for automatic 3D segmentation of brain MR images.

Keywords: SPECT, PET, MRI, 3D Segmentation, Deep Learning
Poster panel: 388

Poster Number:

Auto-classification of respiratory trace using convolutional neural network for adaptive respiratory gated myocardial perfusion SPECT (#1931)

G. S. Mok1, 3, Q. Zhang1, J. Sun1, D. Zhang1, P. H. Pretorius2, M. A. King2

1 University of Macau, Department of Electrical and Computer, Taipa, China, Macao Special Administrative Region
2 University of Massachusetts Medical School, Department of Radiology, Worcester, Massachusetts, United States of America
3 University of Macau, Faculty of Health Sciences, Taipa, China, Macao Special Administrative Region


Previously we showed that respiratory patterns affect the effectiveness of respiratory gated SPECT. This study aims to use a convolutional neural network (CNN) to automatically classify respiration with apnea (AR) and regular respiration (RR) for selecting appropriate rebinning schemes in respiratory gated myocardial perfusion SPECT. We reviewed and classified respiratory traces from 1000 patients tracked from Vicon Motion Systems during their routine 99mTc-MIBI stress SPECT/CT scans. The traces were first pre-processed and a total of 25,000 data points were used for analysis. All traces were first classified into AR and RR by visual inspection. We randomly chose 700 and 300 processed signals for training and testing our proposed CNN, respectively. Selected SPECT data for the classified patients were then rebinned into 7 respiratory gates using phase and amplitude-based gating accordingly. Estimated motion amplitude, FWHM of the image profiles across the left ventricle with Gaussian fit, and normalized standard deviation (NSD) of a uniform region in lungs in different gates were measured from the two rebinning methods for two classified respiratory traces. Our results show that the binary CNN classification accuracy reaches 88%. The mean estimated motion amplitudes are 0.67 cm and 0.94 cm for phase gating and amplitude gating respectively for RR while they are 0.35 cm and 1.66 cm for AR correspondingly. The FWHM of inferior left ventricle are 13.57 mm and 13.07 mm for phase gating and amplitude gating respectively for RR, while they are 15.58 mm and 13.99 mm for AR. The NSD values are about 40% lower for phase gating as compared to amplitude gating respectively for both traces. The use of CNN can classify AR and RR with high accuracy which can be used to guide the subsequent rebinning schemes. Amplitude-based gating is more suitable for AR patients while RR patients are less sensitive to different gating methods, and phase gating could be used to lower noise.

Keywords: Myocardial Perfusion, SPECT/CT, Respiratry gating, Convolutional Neural Network
Poster panel: 391

Poster Number:

A deep learning approach to quantification of PET tracer uptake in small tumours. (#2623)

L. Dal Toso1, E. Pfaehler2, R. Boellaard2, 3, J. Schnabel1, P. Marsden1

1 King's College London, School of Biomedical Engineering & Imaging Sciences, London, United Kingdom
2 University of Groningen, Department of Nuclear Medicine and Molecular Imaging, University Medical Center Groningen, Groningen, Netherlands
3 VU University Medical Center, Department of Radiology and Nuclear Medicine, Amsterdam, Netherlands


The standardised uptake value (SUV) and related methods are widely used for quantification of tumour radiotracer uptake in Positron Emission Tomography (PET) imaging. They are used for evaluation of lesion malignancy, staging and monitoring tumour response to treatment. However, the accuracy of such indices is limited by effects related to the image acquisition process, notably spatial resolution and noise. Improved quantification is expected to lead to more appropriate disease management. To achieve this we have developed a deep learning approach with the aim of more accurately quantifying tumour uptake in PET studies. A 3D CNN is trained on sets of simulated tumour tracer uptake distributions (‘ground truth’) and corresponding simulated PET images. When presented with an unseen simulated PET image the network provides an improved estimate of the corresponding ground truth. The ground truth data used for training are created by generating 3D shapes with an assigned activity distribution. These shapes are then warped in different ways to augment the dataset. To simulate the PET data acquisition and reconstruction process, Gaussian blurring and noise are applied to the ground truth images, producing a complementary dataset comprising simulated PET images. Tumour uptake distributions predicted by the CNN yield an improved estimate of the activity distribution. We conclude that this method leads to a better quantification of tumour radiotracer uptake in simulated data. Preliminary tests of the process applied to physical phantom data show promising results in terms of the recovery of tumour shape and activity concentration.

Keywords: Convolutional neural network, PET, Quantification
Poster panel: 394

Poster Number:

Acquiring and Processing of Light Field Images in a Tissue Equalent Scintillator Based on 3D Dosimetry (#1460)

X. Yan1, H. Li1, L. He1, 2, D. Li1, X. Dai1, X. Zhang1, 3

1 China Institute for Radiation Protection, Taiyuan, China
2 Tsinghua University, Beijing, China
3 North China Electric Power University, Beijing, China


In recent years, the 3D dose measurement technology based on light field imaging of scintillation light in the field of dynamic radiation therapy has developed rapidly. The previous research results showed that the differences of mass attenuation coefficient of the gel and the recommended value of the soft tissue given by ICRU for γ-rays are about 10% for Eγ <150 keV and 0.1% for Eγ >150 keV. And the light yield of the developed gel scintillator can reach about 50% of the one of the standard liquid scintillators. Additionally, tolerance experiments and transmittance measurement experiments on various shell materials have been completed, and the phantom made of scintillation gel has also been formed. The purpose of this work is to research the acquisition and processing of light field images. Firstly, an ultra-low noise & high sensitive light field camera which spatial resolution can reach the sub-millimeter or even micron level is built by using a micro-lens array. And the point spread functions (PSF) of the light field camera on different refocus positions are also studied by knife edge method. Secondly, the 3D information of the object is reconstructed after the depth information recorded in the light field image is extracted. As exploration, reconstructing the 3D information of the inclined ruler is carried out, and the result which the length of the reconstructed inclined portion compared with the actual scale shows an error of about 5%. In addition, for the reconstruction of the internal scintillation light in the transparent scintillation gel, the optical layered imaging experiment using the candle flame as the research object is carried out, and the actual light intensity distributions of the candle flame at different levels are about to be obtained. The research in this paper will be continued, and the depth acquisition algorithm and the refocusing algorithm will be continuously optimized to obtain the better 3D reconstruction results.

Keywords: Radiation Therapy, 3D Dose Measurement, Scintillation Light, Light Field Imaging, Image Reconstruction
Poster panel: 397

Poster Number:

First demonstration of portable Compton camera to visualize 223-Ra concentration for radionuclide therapy (#2083)

K. Fujieda1, J. Kataoka1, S. Mochizuki1, L. Tagawa1, S. Sato1, R. Tanaka1, K. Matsunaga2, T. Kamiya2, T. Watabe2, H. Kato2, E. Shimosegawa2, J. Hatazawa2

1 Waseda University, Graduate School of Advanced Science and Engineering, Tokyo, Japan
2 Osaka University, Graduate School of Medicine, Osaka, Japan


Radionuclide therapy (RNT) is an internal radiation therapy that can selectively kill cancer cells. Recently, the use of alpha-emitting radionuclides was initiated in RNT owing to their high linear energy transfer (LET) and short range. In particular, 223-Ra is widely used for treating bone metastasis of prostate cancer. Despite its potential for clinical applications, it is difficult to determine whether these alpha-emitting radionuclides have been properly delivered to the target lesion. Therefore, we propose a new method of monitoring nuclear gamma rays promptly and simultaneously, that are emitted from 223-Ra as alpha decay, using a high-sensitivity Compton camera. We first observed a small bottle of 223-Ra (0.56 MBq), the reconstructed image of which converged at the correct position with a spatial resolution of ~20 mm at a plane 10 cm in front of the camera. Next, we observed a phantom consisting of three spheres with diameters ranging from 13 to 37 mm filled with 223-Ra solution (9 kBq/mL) and surrounded by a ~20-cm layer of water. A three-dimensional (3D) image was constructed by rotating the Compton camera around the phantom. Images were acquired from eight directions at 30 min intervals. Although the image resolution remained limited, three spheres were resolved at the correct position in the 3D image with their relative intensities. Lastly, we observed the body of a patient for 10 min and the reconstructed image was similar to that acquired by a single-photon emission computed tomography (SPECT) system for 30 min. While the spatial resolution of the Compton camera was worse than the SPECT system, the Compton camera obtained a wider image in a shorter time. In conclusion, we discuss current problems and a new camera we have been developing with improved sensitivity and angular resolution for future clinical applications.

Keywords: Compton camera, MPPC, Radionuclide therapy
Poster panel: 400

Poster Number:

supported by: