Please note! All times in the online program are given in New York - America (GMT -04:00) times!

New York - America ()
Dec 5, 2021, 8:28:16 AM
Your time ()
Tokyo - Asia ()
Dec 5, 2021, 10:28:16 PM
Our exhibitors and sponsors – click on name to visit booth:

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

MIC-Poster IV

Shortcut: M-16
Date: Friday, 22 October, 2021, 12:40 PM - 1:30 PM
Room: MIC - 1,MIC - 2
Session type: MIC Session


Click on an contribution to preview the abstract content.


Mouse Brain PET with a Staggered 4-Layer DOI Detector – Optimized Crystal Layer with GATE (#1264)

H. G. Kang1, H. Tashima1, F. Nishikido1, M. Higuchi1, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Department of Advanced Nuclear Medicine Science, Chiba, Japan


For mouse brain dedicated positron emission tomography (PET), depth-of-interaction (DOI) information is necessary to provide high resolution without compromising the sensitivity. In this study, we designed mouse brain PET with a staggered 4-layer DOI detector using 1 mm crystal pitch for high-resolution molecular imaging research. The proposed PET had a 50 mm ring diameter and 11 mm axial FOV. The proof-of-concept experimental results obtained with a single DOI detector showed good crystal map quality, resolving 95% of the crystals. Based on those results, we conducted GATE simulations to optimize the crystal layer configuration which can provide the best spatial resolution throughout the field-of-view. Three different crystal layer configurations were used: (1) uniform crystal layer (5+5+5+5 mm), (2) heterogenous crystal layer (3+3+4+10 mm), and (3) heterogenous crystal layer with a total thickness of 17 mm (3+3+4+7 mm). To evaluate the spatial resolution, five cylindrical sources (D=0.1 mm, L=10 mm) with spacing of 5 mm in the radial direction were used. In addition, an ultra-micro hot phantom was used to evaluate the imaging performance. The PET images were reconstructed by using an OSEM algorithm with 8 subsets and 10 iterations. An energy window of 400-600 keV was used for the image reconstruction and sensitivity calculation. The heterogenous crystal layer configuration (3+3+4+7 mm) showed the best spatial resolution (0.45 mm at center and 0.74 mm at 10 mm radial offset) among the three. The 0.75 mm rods of the ultra-micro hot phantom could be resolved with the peak-to-valley ratio of 1.57. In conclusion, the mouse brain PET with a staggered 4-layer DOI detector using the heterogenous crystal layer configuration (3+3+4+7 mm) can provide submillimeter resolution and 1.35% sensitivity. In the future, we plan to develop a prototype mouse brain PET scanner to validate the simulation results.

Keywords: GATE, Small animal PET, DOI

Evaluation of a PET detector for a next generation preclinical PET/EPRI (#155)

H. Kim1, Y. Hua4, C. - T. Chen1, Q. Xie3, B. Epel2, S. Sundramoorthy2, H. Halpern2, C. - M. Kao1

1 University of Chicago, Department of Radiology, Chicago, Illinois, United States of America
2 University of Chicago, Department of Radiation and Cellular Oncology, Chciago, Illinois, United States of America
3 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
4 Raycan Technology Co, Ltd., Suzhou, China


We are developing a next generation PET/EPRI hybrid scanner, based on the encouraging animal imaging results produced by PET/EPRI imager. The new PET detector will have ~10.0 cm axial field-of-view, which is 4 times longer than the present system to provide full body imaging for small animals, and thereby enabling PET and EPRI correlation studies of tumors, particularly relating to hypoxia. In addition to the longer axial field-of-view, smaller crystals (pitch ~1.0 mm) will be used to achieve sub-millimeter spatial resolution required for small animal imaging. We devised a new method for SiPM signal readout by combining the traditional and the strip-line based multiplexing to efficiently handle the increased number of SiPMs in the new PET. A prototype PET detector module was built by adopting the newly devised readout method, and its various performance properties were measured. Preliminary results shows that the prototype PET detector is suitable for the next generation PET/EPRI. We present the design of the PET detector, and preliminary results obtained from the performance test.

Keywords: Positron Emission Tomography, Electron Paramagnetic Resonance Imaging, Strip-line readout

Three dimensional (3D) optical imaging of muon beams using a plastic scintillator plate (#203)

S. Yamamoto1, K. Ninomiya2, N. Kawamura3, 4, Y. Hirano1

1 Nagoya University, Graduate School of Medicine, Nagoya, Japan
2 Osaka University, Osaka, Japan
3 High Energy Accelerator Research Organization (KEK), Tokai, Japan
4 J-PARC Center, Tokai, Japan


Although optical imaging of muon beam is a promising method for range and width estimations, it was limited only two dimensional (2D) projection images. For the beam shape determinations of muon beams, 3 dimensional (3D) beam shape distribution is desired. To measure the 3D beam shape distribution, we conducted optical imaging of muon beam using a plastic scintillator plate set in water phantom. A plastic scintillator plate was dipped in a water phantom and positive muon beam was irradiated from the alongside of the scintillator plate at J-PARC. The optical image in the scintillator plate was acquired using a charge-coupled device (CCD) camera during irradiation from the side. The imaging system was moved 10 mm steps perpendicular to the beam direction to acquire the set of sliced optical images of the beam. The sliced optical images were stacked, interpolated to form a 3D optical image and the depth and lateral profiles were evaluated. From the depth profile derived from the 3D images, the Bragg peak position was estimated precisely. The lateral profiles at Bragg peak could also be derived. We confirmed that 3D optical imaging of muons was possible and is a promising method for the beam shape distribution, research of muons, as well as the future muon radiotherapy.

Keywords: muon, optical imaging, plastic scintillator plate, 3D

A Four-Layer DOI PET Detector Using 1 mm Crystal Bars Segmented by Subsurface Laser Engraving (#319)

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

1 National Institutes for Quantum and Radiological Science and Technology (QST), Chiba, Japan


The spatial resolution of positron emission tomography (PET) scanners is improved significantly by depth of interaction (DOI) information. Therefore a four-layer DOI detector has been developed in our group based on light sharing between small discrete crystals in each layer by the specific reflector configuration. However, the precise fabrication of segmented crystal arrays is crucial and time consuming. In the current study, we investigated the performance of our four-layer DOI detector composed of crystal bars segmented into four segments by applying the subsurface laser engraving (SSLE) technique. The LYSO crystal bars of 1×1×20 mm3 were segmented in the height direction using SSLE into 4 segments and an 8×8 crystal array was prepared. We compared the performance of the proposed detector with a similar detector composed of monolithic crystal bars and our conventional four-layer detector consisting of discrete crystals of 1×1×5 mm3 from the viewpoints of the crystal identification and energy resolution. The 2D position maps of the detectors were obtained by the Anger type calculation and the crystal identification performance was evaluated for the crystal arrays. Clear separation was obtained for the proposed detector at the middle of the crystal array. The average energy resolution of 8.2% ± 1.0% at 662 keV was obtained for the proposed detector, which was a 20% improvement compared to the conventional four-layer DOI detector.

Keywords: four-layer DOI detector, segmented crystal bars

A Single-Line Multi-Voltage Threshold Method to Improve the Integrity of FPGA-Only Data Acquisition System (#342)

H. Chu1, 2, M. Yi3, J. S. Lee1, 3

1 Seoul National University, Department of Nuclear Medicine, Seoul, Republic of Korea
2 Ulsan National Institute of Science and Technology, Department of Electrical and Computer Engineering, Ulsan, Republic of Korea
3 Seoul National University, Interdisciplinary Program in Bioengineering, Seoul, Republic of Korea


Multi-voltage threshold (MVT) method requires multiple comparators and timing channels. Therefore, when implementing the MVT method using FPGA-based time-to-digital convertors, the limited number of I/O ports in the FPGA restricts the number of analog signals that can be processed. Therefore, in this study, we propose a new single-line MVT method to improve the integrity of FPGA-only data acquisition system by reducing the FGPA input channels required for the MVT method. The single-line MVT method alleviates the channel number problem of the original MVT method by employing an XOR gate that integrates all digital output signals from comparators and provides 1-bit digital pulse train. Because the proposed method allows only the odd number of digital pulses as input to the XOR gates, we have examined the feasibility of using only 3 thresholds for energy estimation and signal reconstruction in the MVT method. In addition, the circuit design and simulation of the single-line MVT was performed using LT-SPICE program. MVT with 3 thresholds showed good energy correlation (R2 = 0.96). Signal reconstruction accuracy was 0.18 in NRMSE when applying mixed Gaussian curve fittings. Circuit simulations have shown that all 6 time stamps at 3 thresholds were well detected in 1-bit pulse trains, providing timing and energy information. The reduction of timing channels for MVT by applying the proposed method will be useful in improving the integrity of PET data acquisition system. The experimental results will be presented at the conference.

Keywords: Positron Emission Tomography (PET), time-over-threshold (TOT), multi-voltage threshold (MVT) method

Pushing the Timing Limits of the TOFPET2 ASIC (#394)

V. Nadig1, S. Gundacker1, H. Radermacher1, D. Schug1, 2, B. Weissler1, 2, L. Yin1, R. Allendorf1, V. Schulz1, 2

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


With the major goal of reaching a coincidence time resolution (CTR) well below 100 ps, not only scintillators and photosensors, but also front-end electronics have become the focus of latest efforts to improve the CTR in time-of-flight positron emission tomography (TOFPET). This study adapted a high-frequency (HF), baseline-shift compensating front-end circuit combined with an ASIC by PETsys Electronics S.A. to initiate the transfer of this benchtop readout technique to PET systems. Testing Hamamatsu silicon-photomultipliers (SiPMs) coupled to 3 mm x 3 mm x 20 mm Cerium-doped lutetium-oxyorthosilicate (LYSO:Ce) crystals with the combined HF and TOFPET2 ASIC readout architecture improved the CTR to 191 ps compared to a reference CTR of 200 ps acquired with the PETsys TOFPET2 Evaluation Kit. For Broadcom SiPMs and the same crystals, a CTR of 186 ps (HF) compared to 202 ps (reference) is reported. Smaller LYSO:Ce crystals of 2 mm x 2 mm x 3 mm size coupled to the same SiPMs reached CTRs of 128 ps and 135 ps (HF) compared to 135 ps and 149 ps (reference). Despite the significant improvement of the CTR by the HF readout architecture, the amplified signals provoked up to four side peaks in the coincidence time difference spectra with a periodic distance of 350 ps to 390 ps to the main peak. In-depth studies revealed that these side peaks are also visible within standard Evaluation Kit measurements. It is shown that the peak amplitude is reduced with decreased SiPM signal height and by re-configuring the trigger logic to a single-threshold trigger in the energy branch.

AcknowledgmentThe authors would like to thank Stefan Brunner from Broadcom for providing samples and Stefan Tavernier, Ricardo Bugalho and Luis Ferramacho from PETsys Electronics S.A. for sharing their expertise in our many discussions.
Keywords: time-of-flight, PET, high-frequency readout, CTR

Reading a Hexagonal Matrix of SiPMs With a Dedicated ASIC (#474)

R. Chil1, D. Pérez-Benito1, L.A. Hidalgo-Torrez1, J.J. Vaquero1, 2

1 Universidad Carlos III de Madrid, Departamento de Bioingeniería e Ingeniería Aeroespacial, Leganes, Spain
2 Instituto de Investigación Sanitaria Gregorio Marañón, Madrid, Spain


We present a comparative study to test the feasibility of using a dedicated ASIC to read a radiation detector for small animal PET imaging as an alternative for conventional approaches based on analog multiplexing and ADCs, aiming at high spatial and temporal resolution. The SiPM detector hexagonal array side is 17.61mm, comprised of 61 individual hexagonal channels, with a side of 2.25 mm each, and it is manufactured by Hamamatsu (Japan). This detector is coupled to a squared scintillator array  with a pixel pitch of 1.28mm. We compare two setups; setup A: analog multiplexing going from 61 to 4 output channels acquired with a 125MHz ADC; setup B: 61 channels connected directly to a Triroc ASIC (Weeroc, France). We obtained a pixel resolvability index of 0.32 for setup A and 0.16 for setup B, proving that both systems can work with this detector/scintillator configuration and that setup B offers a better solution.

AcknowledgmentThis research has received funding from the project ERA4TB, Innovative Medicines Initiative 2 Joint Undertaking (JU) under grant agreement No 853989. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and EFPIA and Global Alliance for TB Drug Development non profit organisation, Bill & Melinda Gates Foundation and University of Dundee.
This work was partially funded by project PID2019-109820RB-I00 from the Spanish Ministry of Science and Innovation.
Keywords: PET, detector, ASIC, hexagonal, SiPM

Improved edge crystal separation in the 4-layer DOI PET Detector with a staggered reflector arrangement (#705)

N. Inadama1, S. Takyu1, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology, Chiba, Japan

This work was supported in part by the Japan Society for the Promotion of Science through Grant-in-Aid for Scientific Research under Grant 20K12705.


We have previously developed a 4-layer DOI PET detector based on our original reflector arrangement, in which 3D crystal location can be identified by the 2D Anger-type calculation. Even with a resistor chain for a photo-detector, the DOI detector has the ability to identify most of the 32 × 32 × 4 crystals with the signals reduced to 16 channels; however, identification of outer crystals has long been an unsolved issue. The problem comes from the fact that scintillation light from the outermost crystal spreads only in the inner direction. In this study, we proposed an improved reflector pattern for better identification of the outermost crystals. The key factor was the slits which are typically formed in reflectors to assemble a reflector lattice.

Keywords: PET detector, DOI detector

Timing resolution of TlBr and TlBr-TlCl PET detectors based on Cerenkov radiation measurement (#794)

F. Nishikido1, K. Hitomi2, M. Nogami2, H. G. Kang1, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Chiba, Japan
2 Tohoku University, Aomori, Japan


TlBr is a high density (7.56 g/cm) semiconductor material composed of high effective atomic number elements. Therefore, TlBr has sufficient detection efficiency for 511 keV annihilation radiation in PET. In addition, TlBr detectors have high energy resolution due to the direct conversion to electrons. On the other hand, timing performance of typical semiconductor detectors is poor compared with scintillation detectors However, detection of Cerenkov light from TlBr offers an alternative method to get better timing resolution. This paper reports our feasibility study on the timing performance of TlBr detectors using Cerenkov light.

In the experiment, a conventional TlBr crystal and TlBr mixed with TlCl (TlBr-TlCl crystal) were used. The TlBr crystals were fabricated at Tohoku University. Each TlBr crystal was 3 × 3 × 3 mm3. The TlBr crystals were covered with ESR films and Teflon tape. Cerenkov light was detected with a multi-pixel photon counter (Hamamatsu S13360-3075CS). Coincidence detection measurement was carried out using a LYSO scintillator coupled with another MPPC (Hamamatsu S13360-3050CS) as a reference detector. Signals from the detectors were amplified with high frequency amplifiers and then waveforms were recorded with a digitizer (CAEN, DT5742).

We obtained the timing spectra by Cerenkov light from the TlBr and scintillation light from the LYSO. From them, the timing resolution better than 600 ps was obtained for all the event data. After selecting the events by optimizing a trigger level, the timing resolutions better than 450 ps was obtained for both the TlBr detector and the TlBr-TlCl detector.
Keywords: TlBr, TOF-PET, PET detector, Cerenkov radiation

Towards 200 ps CRT in DOI-capable Semi-Monolithic PET-Detectors for Clinical Applications (#938)

S. Naunheim1, T. Solf2, Y. Kuhl1, D. Schug1, 3, V. Schulz1, 3, F. Mueller1

1 RWTH Aachen University, Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, Aachen, North Rhine-Westphalia, Germany
2 Philips Digital Photon Counting (PDPC), Aachen, North Rhine-Westphalia, Germany
3 Hyperion Hybrid Imaging Systems GmbH, Aachen, North Rhine-Westphalia, Germany


In PET imaging, timing performance is usually evaluated by coincidence resolving time (CRT) representing a key characteristic of PET detectors. Good timing resolution enables the possibility using time-of-flight information, thus improving the image’s signal-to-noise ratio. Semi-monolithic detectors provide intrinsic depth of interaction (DOI) capabilities, and combine advantages of segmented and monolithic detectors, namely small read-out area and high density of scintillation photons. The used coincidence measurement setup consists of a semi-monolithic slab detector comprising 8 monolithic LYSO slabs (each 3.9 x 31.9 x 19.0 mm3) and an one-to-one coupled detector (64 segments à 3.9 x 3.9 x 19.0 mm3). Both detectors are coupled to 16 digital SiPMs (DPC 3200-22, Philips Digital Photon Counting).
We present a novel timing calibration scheme for semi-monolithic detectors, utilizing multiple modular sub-calibrations, aiming to correct electrical and crystal-related time skews. Each sub-calibration is subsequently applied on the previous one and addresses a different kind of time skew. For calibration and evaluation events within an energy window of 411 keV to 561 keV are used. After calibration a maxed out CRT of 225 ps (combined) and 204 ps (DOI-layer close to the crystal’s top surface) is obtained. Without using specific filters, CRT values of 235 ps (combined) and 215 ps (DOI-layer close to the crystal’s top surface) can be achieved by preserving high sensitivity using all events.

Keywords: Time Calibration, Slabs, Time of Flight, Monolithic Scintillators, Coincidence Time Resolution

Dynamic Time-Over-Threshold Readout for Improved Energy Linearity in PET Detectors that Achieve 100 ps CTR (#1143)

S. Pourashraf1, A. Gonzalez-Montoro1, J. W. Cates2, Z. Zhao3, J. Y. Won4, J. S. Lee4, C. S. Levin5, 6

1 Stanford University, Radiology, Stanford, California, United States of America
2 Lawrence Berkeley National Laboratory, Applied Nuclear Physics Program, Berkeley, California, United States of America
3 Shanghai Jiao Tong University, Biomedical Engineering, Shanghai, China
4 Seoul National University, Nuclear Medicine and Biomedical Sciences, Seoul, Republic of Korea
5 Stanford University, Molecular Imaging Program, Stanford, California, United States of America
6 Stanford University, Radiology, Bioengineering, Physics, and Electrical Engineering, Stanford, California, United States of America


We have embedded a dynamic time-over-threshold (DynTOT) block in our scalable TOF-PET detector readout electronics to linearize the multiplexed energy spectra while keeping 100 ps FWHM Coincidence Time Resolution (CTR) performance. This DynTOT block is constructed by off-the-shelve components and consumes only 20 mW power per detector ‘layer’ in our design. Using 3×3×10 mm³ “LGSO” crystals coupled to arrays of 3×3 mm² SiPMs, energy resolution of 13.8±0.1%, 13.9±0.3%, and 17.1±0.6% were experimentally achieved for conventional pulse height, DynTOT, and conventional TOT methods, respectively. Including the DynTOT method, coincidence data were experimentally acquired using a low jitter FPGA-based TDC, and 102.2 ± 1.3 ps FWHM CTR was achieved, demonstrating the robustness of our novel and scalable timing electronic chain.

AcknowledgmentThis work was supported in part by NIH research grants 5R01CA21466903, 1R01EB02512501, and by NRF- 2016R1A2B3014645 National Research Foundation of Korea. Andrea Gonzalez- Montoro is partially supported by VALi+d Program for Researchers in Postdoctoral Phase of the Ministry of Labor and Social Economy (Generalitat de Valencia) and the European Social Fund. We also thank Xilinx University Program, providing us with Kintex- 7 FPGA kit and associated licenses. In addition, we would like to thank Mr. Takeyama Toshinori- NYKSCHM, Marubeni America Corporation, and Oxide Corporation for providing fast LGSO scintillation crystals.
Keywords: TOF-PET, energy Linearity, Energy Resolution, Dynamic TOT, 100 ps CTR

Design of a Novel DOI, TOF, High Spatial Resolution Detector for Human Brain PET (#1206)

W. He1, J. Wang1, X. Zhao1, Y. Zhao1, D. Prout2, A. Chatziioannou2, Q. Ren1, Z. Gu1

1 Shenzhen Bay Laboratory, Institute of Biomedical Engineering, Shenzhen, China
2 University of California, Los Angeles, Crump Institute for Molecular Imaging, Los Angeles, California, United States of America


A novel detector design is proposed for high sensitivity and high resolution human brain PET imaging. The aim is to simultaneously achieve ≥4 levels (≤ 7 mm) DOI resolution, ~1 mm spatial resolution, and ≤300 ps TOF timing resolution, while maximizing sensitivity. Monte Carlo simulations were performed to show the feasibility of the DOI decoding method. Based on the proposed detector, a preliminary brain PET system was designed and the system sensitivity was evaluated.


The detector features a multi-resolution design and is comprised of two layers of LYSO crystal arrays. The top layer is a 20×20 array of 1.3×1.3×6 mm3 LYSO crystals, aiming to provide high spatial resolution. The bottom layer is an 8×8 array of 3.22×3.22×20 mm3 LYSO crystals one-to-one coupled to an 8×8 SiPM array, aiming to achieve high TOF timing resolution. A 2 mm thick glass lightguide is placed between the top and the bottom LYSO layers, allowing high DOI resolution. The preliminary brain-dedicated PET system has 16 rings each comprised of 32 blocks, forming a bore diameter of 30 cm and an axial length of 40 cm.


Geant4 simulation of scintillation light transport indicates that the proposed detector has the potential of decoding at least four- levels of DOI using pmax/p, where pmax is the maximum value of all SiPM channels and p is the summation of all SiPM signals. Furthermore, a better DOI resolution can be achieved by applying crystal location dependent pmax/p thresholds. Simulations show that diffused reflector should be used in the bottom LYSO layer to enable DOI decoding, and a thicker lightguide can further improve DOI resolution. With an estimated TOF timing resolution, the effective sensitivity of the proposed brain-dedicated PET system is 33.9%.


In conclusion, the preliminary results demonstrated the potential of a novel DOI, TOF, high spatial resolution detector for high sensitivity and high resolution human brain PET.

Keywords: brain-dedicated PET, DOI, TOF

Removal of electronegative impurities andcharacterization of charge read-out in TMBi (#1285)

S. Peters1, B. Gerke2, K. Bolwin2, V. Hannen1, C. Huhmann1, N. Markwardt1, K. P. Schäfers2, C. Weinheimer1

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


A new type of detector for positron-emission tomography (PET) has been proposed recently, using a heavy organo-metallic liquid - TriMethyl Bismuth (TMBi) - as target material. TMBi is a transparent liquid with the high-Z element Bismuth possessing 82% of its mass. Thus the  511keV annihilation quanta are converted efficiently into photo-electrons within the detector material producing both Cherenkov light and free charge carriers within the liquid. While the optical component enables a fast timing, a charge readout using a segmented anode can provide an accurate position reconstruction within the detector. The charge measurement requires a high level of purification, as any electronegative contaminants can cause signal degradation and will produce noise within the detector. In addition to the purity requirements, the reactive nature of TMBi poses many challenges that need to be met until a fully functioning detector for PET applications can be realized. Here a new setup is presented that allows to remove electronegative impurities and to characterise properties of TMBi w.r.t. its application as a detector medium for charge read out.

AcknowledgmentThis work is supported by the DFG, Project Number: WE 1843/8-1; SCHA 1447/3-1
Keywords: PET, TMBi, purification, detector

Experimental study on the light output of the photonic crystals fabricated using femtosecond laser (#1373)

X. Yu1, X. Zhang1, S. Xie2, J. Xu1, Q. Peng2

1 Huazhong University of Science and Technology, Wuhan, China
2 Shenzhen Bay Laboratory, Shenzhen, China


One of the keys to improve PET imaging performance is to improve light output of the detector module. At present, the scintillation crystal used in PET detectors cannot output satisfactory amount of intercepted light because the high refractive index causes the phenomenon of total reflection, which seriously affects the imaging effect. Femtosecond laser is a promising method to process a large number of photonic crystal structures on conventional scintillation crystals to improve light output.

Following conclusion can be drawn: (1) Femtosecond laser can be implemented to fabricate photonic crystal. The parameters of the laser may have great impact on the performance of light output. (2) Initial experiment showed that the micro-structure can greatly improve the light output by over 50%. (3) Simulation using Tracepro matched perfectly to the experimental result.

The next step is to determine the appropriate process parameters and processing methods. Techniques of processing practical large-scale surface will be studied, too.
AcknowledgmentThis work was supported by the National Natural Science Foundation of China (51627807), and Shenzhen Science and Technology Program (KQTD2017033015530192). We sincerely thank the Micro and Nano Fabrication and Measurement Laboratory of Mechanical Science and Engineering in HUST.
Keywords: light output, photonic crystal, femtosecond laser

Initial Evaluation of a TOF, Open-ring PET Scanner with Continuous Variable Diameter (#42)

B. Li1, L. Fang2, B. Zhang2, L. Yang3, P. Xiao2, 5, X. Chen4, Q. Xie2, 5

1 University of science and technology of China, Institute of Artificial Intelligence, Hefei, China
2 Huazhong University of Science and Technology, Biomedical Engineering Department, Wuhan, China
3 Hubei RaySolution Digital Medical Imaging Technology Co., Ltd.,, Ezhou, China
4 Chuxiong Normal University, School of Math and Compter Science, Chuxiong, China
5 University of science and technology of China, School of Information Science and Technology, Hefei, China


We propose a TOF, open-ring PET scanner with continuously variable diameter for in-situ imaging. The detectors of the scanner could be spread or shrunk along circumference direction, provides an adaptive FOV for bodies of different sizes. The annular shape contributes to higher sensitivity and better response homogeneity than the panel scanner composed with the same detector number and internal diameter. The simulation results show the image quality and the response homogeneity of the open-ring scanners get worse than those of the full-ring scanner (with coincidence time resolution of 380 ps), but the fluctuation of the background variance (improved 4.5%∼6.1%) and the hot lesion contrast (reduced 3.9%∼11.8%) is acceptable even the open angle reaches up to 180◦.

AcknowledgmentThis work was jointly supported by the National Key Research and Development Program of China #2019YFC0118900, and the National Research and Development Program for Major Research Instruments of NSFC of China #61927801 and #62027808.
Keywords: time of flight (TOF), all-digital PET, image reconstruction, open-ring PET

A Monolithic Crystal PET Detector Using Integrated Corner Retroreflectors (#221)

L. Saleh1, P. Vaska1, 2

1 Stony Brook University, Biomedical Engineering, Stony Brook, New York, United States of America
2 Renaissance School of Medicine at Stony Brook University, Radiology, Stony Brook, New York, United States of America


We investigate the potential improvements in PET imaging performance from machining integrated corner retroreflectors (ICR) directly onto the front face of monolithic scintillators to improve spatial resolution, light collection, and depth of interaction capability. Standard PET detectors use small pixelated scintillators to confine light which results in significant inter-crystal scattering and dead space for interspersed reflectors. Monolithic scintillators do not have dead space and are easier to manufacture, yet spread light over a larger detection area. We use a trihedral corner reflector as used in retroreflectors, which efficiently reflect light back to its origin and readout by silicon photomultiplier (SiPM) array.  Similar earlier approaches involved optical coupling of a separate retroreflector structure, which introduces additional scattering and inefficiency.  We used GATE, a Monte Carlo toolkit to simulate a 10 mm thick monolithic scintillation crystal with different entrance face finishes: fully absorptive (black), flat with a polished specular reflector, and machined 0.1 mm corner reflectors with a polished surface coupled to a specular reflector. The ICR surface reduce the width of the light spread function substantially. With no readout noise, the ICR surface improves transverse spatial resolution by ~30% relative to the fully absorptive and flat specular surfaces. When adding a more realistic readout, the ICR detector still results in an improvement in resolution >20%.  While the depth of interaction resolution is similar across methods at depths near the SiPM while the corner reflector is far superior at shallow depths.


We thank Stony Brook University Graduate School and the Dr. W. Burghardt Turner Fellowship for the support.

Keywords: Positron Emission Tomography (PET), Monolithic Crystal, Depth of Interaction (DOI)

Introduction of Spread Field Imaging—a Novel High Performance Collimation for SPECT (#311)

Z. Mu1, Z. Tao1, F. Fahey1

1 Argospect Technologies, Foster City, California, United States of America


Collimation plays a vital role in single photon emission computed tomography systems. Spread field imaging is a class of high-sensitivity, high-resolution collimation and imaging technologies that utilizes shared photon paths to improve collimator performance. In this article we introduce a novel design, called spread field imaging-narrow angle , that shares some characteristics of both parallel-hole and pinhole collimators. In this approach, the holes are long and narrow, similar to the holes in a parallel hole collimator; however, as opposed to parallel-hole collimators, there is space between the holes and the detector. With that spacing, the overall thickness of the collimator is around 7-11 cm. In addition, the holes are not identical, possibly varying in shape, size, etc., and may even include closed holes. The hole-detector spacing provides an extended “virtual” hole length that improves resolution.  Holes with wider acceptance angle may be used to achieve higher resolution and sensitivity compared with parallel-hole collimators. An ordered subset expectation and maximization based algorithm has been developed to reconstruct the object image.  Simulations were developed with the three-dimensional Hoffman brain phantom to compare reconstructed results obtained with from spread field imaging collimation with conventional low-energy, high-resolution parallel hole collimators with full, half or a quarter of conventional imaging times. The results showed that the spread field imaging-narrow angle collimator yielded better or similar reconstruction compared with low-energy, high-resolution even when the imaging time was cut to a quarter of the conventional time. These results demonstrate that spread field imaging-narrow angle collimators have the potential to be high-resolution, low-dose collimators applicable to conventional single photon emission computed tomography systems, and suitable for a wide range of clinical applications.

Keywords: SPECT, collimator, Spread Field Imaging, OSEM, high-sensitivity high-resolution

Addition of Cerenkov LUTs in the LUTDavisModel for optical Monte Carlo simulation in GATE/Geant4 (#435)

E. Moghe1, C. Trigila1, E. Roncali1, 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


Cerenkov photons are optical photons promptly generated by recoil electrons ejected by photoelectric or Compton effect within a scintillator. Cerenkov photons are of great interest and could improve the coincidence timing resolution in Positron Emission Tomography (PET). Differences in emission spectrum and polarization change the fate of Cerenkov and Scintillation photons at the crystal boundaries, determined through the Fresnel equations. Previously, our group developed an algorithm to compute the reflectance, transmittance, and angular distribution of reflected and transmitted photons in scintillation crystals, based on a 3D scan of the crystal surface. To implement the algorithm, our group also developed a standalone which creates a Scintillation photons’ LUT by defining the surface, crystal, and coupling medium. The LUTs can then be used on optical Monte Carlo simulations of scintillation detectors used in PET systems. The goal of this work is to integrate Cerenkov photons’ characteristics within the standalone application to compute Cerenkov LUTs. The users can choose between Scintillation and Cerenkov photons before generating a LUT. These LUTs can be used to study the Cerenkov transport and detection within a scintillation material by performing Monte Carlo simulation using GATE. Since separate LUTs are generated for each photon type, we created an additional panel that could merge the Scintillation and Cerenkov LUTs to study them simultaneously. In this work, we present the integration of Cerenkov photons’ characteristics within the standalone application, highlighting the changes from the original version.

AcknowledgmentThis work was supported by NIH grant R01EB027130. E. Moghe and C. Trigila are with the Department of Biomedical Engineering, University of California Davis, Davis, CA 95616, USA. E. Roncali is with the Department of Biomedical Engineering and the Department of Radiology, University of California Davis, Davis, CA 95616, USA.
Keywords: Simulations, Modeling, Optical transport, Cerenkov, GATE

Simulation study on the feasibility of Time-over-Threshold based positioning in monolithic PET detectors (#496)

C. Thyssen2, 1, P. Mollet2, K. Deprez2, R. Van Holen2, 1, S. Vandenberghe1

1 Ghent University, Medisip, Gent, Belgium
2 Molecubes NV, Gent, Belgium


In the preclinical MOLECUBES β-CUBE positioning in monolithic LYSO crystals is done based on signal integration. A drawback of this approach is the need for power-hungry and rather bulky discrete components. A possible redesign in system electronics could be based on Time-over-Threshold (ToT) with off-the-shelf components. In this work we use Monte Carlo simulations to study the feasibility of ToT, with thresholds up to 31 p.e. (photoelectron), in our monolithic detectors. In all situations we found a resolution between 0.8 mm and 1.25 mm and the maximal degradation was 0.1 mm. Also the bias showed no remarkable degradation in performance and for points far enough from the detector edges the bias did not exceed 0.2 mm for ToT positioning. For this reason we concluded that a monolithic ToT-based detector is feasible and worth further investigation.

Keywords: Monte Carlo simulations, GATE, spatial resolution, time-over-threshold, PET detector

Ultra-fast Fully Monte Carlo PET Reconstruction (#789)

P. Galve1, F. Arias-Valcayo1, A. Lopez-Montes1, A. Villa-Abaunza1, P. Ibanez1, J. L. Herraiz1, 2, J. M. Udias1, 2

1 Universidad Complutense de Madrid, CEI Moncloa, 1Grupo de Física Nuclear, EMFTEL & IPARCOS, Madrid, Spain
2 Instituto de Investigación Sanitaria del Hospital Clínico San Carlos (IdISSC), Madrid, Spain


The success of image reconstruction in PET depends on the accuracy of the system response matrix (SRM) that models all the physical processes involved. The Ultra-fast Monte Carlo PET simulator (UMC-PET) is an accurate, fast and flexible PET simulator which has been developed for multiple purposes. UMC-PET can be used for iterative image reconstruction with a projection step based on full MC calculations (UMC projection), including positron range, non-collinearity, photon transport (scatter and attenuation), and detector response (depth of interaction, energy resolution, multi-detector and multi-crystal events), and thus avoiding physics simplifications in the SRM. We compare this scheme with traditional projection techniques factorized in geometrical projections, spatially-variant Point Spread Function, attenuation, and scatter, besides possible regularization methods. A faster version of the UMC projection was also studied by exploding the scanner symmetries (symUMC projection) and factoring out scatter estimation and attenuation. The three methods (spatially-variant PSF, UMC projection, and symUMC projection) were assessed in the 6R-SuperArgus preclinical PET/CT scanner, taking a few hours for a full reconstruction for the UMC projection. For the NEMA NU 4-2008 Image Quality phantom, a recovery coefficient up to 40% in the 1 mm rod and a 3.00% spill-over were obtained against a 33.5% recovery and 5.86% spill-over for the spatially variant PSF, in either case at 5% noise level. Noticeable image quality improvement is also seen in a simultaneous acquisition of four rats and a Derenzo phantom. This provides not only a useful and flexible gold standard reconstruction, which can be used for training neural networks, but may become a practical reconstruction approach when combined with variance reduction methods and/or high-performance multi-GPU systems.


FPA2015-65035-P, RTI2018-098868-B-I00, RTC2019-007112-1, TEC2015-73064-EXP, TEC2016-78052-R, B2017/BMD-3888 PRONTO-CM

Keywords: GPU, Positron Emission Tomography, reconstruction algorithms, Monte Carlo simulation

Arc-PET: A High Resolution and High Sensitivity Whole-Body Scanner Based on Prism-PET Modules (#899)

Z. Wang1, X. Zeng2, X. Cao2, E. Petersen1, A. LaBella3, W. Zhao3, A. Goldan3

1 Stony Brook University, Department of Biomedical Engineering, Stony Brook, New York, United States of America
2 Stony Brook University, Department of Electrical Engineering, Stony Brook, New York, United States of America
3 Stony Brook University, Department of Radiology, Stony Brook, New York, United States of America


There is a growing interest in developing a positron emission tomography (PET) scanner with high sensitivity and spatial resolution to enable high-quality dynamic imaging studies with kinetic modeling. Most of the current PET scanners have a cylindrical geometry with a relatively large bore diameter, reducing the geometric efficiency and exacerbating the acollinearity caused image blurring. Here, we propose the Arc-PET, a PET scanner that uses depth-encoding Prism-PET modules to enable non- cylindrical hexagonal geometry that improves the solid angle coverage while reducing the amount of scintillator blocks per ring (a significant cost factor). In this work, we evaluated the performance of the Arc-PET with Monte Carlo simulation by using the GATE simulation toolbox. The NEMA standard was used to characterize the sensitivity, count rate, and spatial resolution of the Arc-PET. A hot-spot phantom with 6 difference rod diameters from 1 mm to 4 mm was placed at both the center and 140 mm away from the center field-of-view (FOV) to evaluate the image quality. The simulation results suggest that the Arc-PET has a peak absolute sensitivity of 24.7% and a maximum noise equivalent count rate (NECR) of 902 Kcps at 18.1 Kbq/cc. Reconstructed images of hot-spot phantom demonstrate that the Arc-PET can reach a uniformly ∼1.5 mm spatial resolution across the entire FOV. Based on our simulation results, the proposed Arc-PET is able to achieve high resolution and high sensitivity while keeping the system cost low.

Keywords: High-resolution positron emission tomography (PET), Prism-PET detector module, DOI-rebinning, Monte-Carlo simulation

Optimization through Monte Carlo Simulations of a novel High-Resolution Brain-PET System based on Resistive Plate Chambers (#1017)

A. L. Lopes1, 2, M. Couceiro3, 2, P. Crespo1, 2, P. Fonte3, 2

1 University of Coimbra, Department of Physics, Coimbra, Portugal
2 LIP - Laboratory of Instrumentation and Experimental Particle Physics, Coimbra, Portugal
3 Politechnic of Coimbra, ISEC/DFM, Coimbra, Portugal


Based on previous results obtained with a pre-clinical PET scanner for mice that uses RPC detectors, which yielded an excellent spatial resolution of 0.4 mm Full Width at Half Maximum, a PET scanner dedicated to brain imaging using RPC detectors, is being developed. This new scanner, named HiRezBrainPET, is expected to provide the spatial resolution needed for brain imaging. The present work comprises the initial studies performed through Monte Carlo simulations towards the optimization of some parameters of the RPC detectors that are being developed for the scanner. The studied parameters include the glass plate thickness that optimizes the detection efficiency for stacks of 4, 8, and 10 RPC detectors in each detection head, and the influence on the detection efficiency of the thicknesses of the materials composing the RPC readout electrodes, both the FR4 of the Printed Circuit Boards and the copper strips used to collect the signal. Also addressed was the fraction of detected coincidences between adjacent detection heads and between opposing detection heads for several point source positions. Considering the results and the costs involved, the glass plate thickness should be equal to 280 μm. Regarding the materials forming the readout electrodes, the one mainly influencing the detector efficiency is the FR4. Also, it is worth implementing a readout system that performs coincidences between all four detection heads.

Keywords: brain-PET, HiRezBrainPET, RPC, simulation

Whole-body PET Monte Carlo simulation by using patient-derived activity and attenuation maps. (#1142)

J. Silva-Rodríguez1, 2, F. J. López-González3, J. Paredes-Pacheco3, P. Aguiar2, 3

1 Health Research Institute of Santiago de Compostela, Nuclear Medicine and Molecular Imaging Group, Santiago de Compostela, Spain
2 University of Santiago de Compostela, Center for Research in Molecular Medicine and Chronic Diseases, Santiago de Compostela, Spain
3 University of Santiago de Compostela, Radiology Department, Faculty of Medicine, Santiago de Compostela, Spain


Previous works have developed methodologies for deriving digital phantoms including physiological variability from patient data. Nevertheless, these methods usually need PET/MR images and are only available for generating brain phantoms. In this work, we present a novel approach for applying this methodology to whole-body PET/CT images. Preliminary maps were derived by translating CT images to linear attenuation coefficients and binning them to discrete tissues. Attenuation was simplified in three tissue types, while activity was calculated for 64 tissue levels. These maps were simulated using a validated GE Discovery ST SimSET model. 5-7 beds were simulated to cover the whole-body phantoms.   Image reconstruction was performed by using STIR. The reconstructed images were compared with the original PET images for generating voxel-wise difference maps that were used to update the original maps. A second simulation was performed with the updated maps. The developed method was able to produce visually realistic images in comparison with computational phantoms. Nevertheless, we found some problems in the map updating process, particularly noise and resolution degradation being translated to the phantoms. Additional improvements are needed in this step before a reliable validation of the method is performed.


JPP and FJLG are funded by Ph.D. scholarships (FPU16/05108 and FPU17/04470, respectively) from the Spanish Ministry of Education, Culture and Sport under the FPU program. PA is a Ramon y Cajal fellow (RYC-2015/17430). This work is partly funded by the NeuroAtlantic project (UE Interreg EAPA_791/2018).

JSR is with the Health Research Institute (IDIS) and with the Center for Research in Molecular Medicine and Chronic Diseases of Santiago de Compostela. JPP, FJLG, and PA are with the Radiology Department of the University of Santiago de Compostela.

Keywords: PET/CT, Monte Carlo simulation, SimPET, STIR

First investigation of list mode MLEM reconstruction for fast DC-SPECT system design optimization (#1196)

Y. Feng1, L. Bläckberg1, G. El Fakhri1, W. Worstell2, H. Sabet1

1 Massachusetts General Hospital & Harvard Medical School, Gordon Center for Medical Imaging, Department of Radiology, Boston, Massachusetts, United States of America
2 PicoRad Imaging LLC, Wayland, Massachusetts, United States of America


A Dynamic Cardiac SPECT (DC-SPECT) system being developed at the Massachusetts General Hospital, Harvard features a cardio focus asymmetrical geometry. In such asymmetrical geometry the life cycle of design optimization from mechanical drawing to Monte Carlo implementation to image reconstruction and evaluation is long and labor intensive. This work presents a first implementation of List Mode Maximum Likelihood Expectation Maximization (LM-MLEM) reconstruction for the DC-SPECT system to enable fast design optimization using analytical method. We investigate the DC-SPECT system imaging resolution by applying a simple system modeling while implementing LM-MLEM. Results show that without any correction of scattering or attenuation effects, LM-MLEM reconstruction can achieve a 5 mm FWHM spatial resolution. The results agree with the expected performance of the DC-SPECT.


The authors acknowledge financial support of the NIH Grant No. R01HL145160.

Keywords: DC-SPECT, list mode MLEM, MC simulation

Investigations of Pinhole Numbers and Projection Views in Multi-pinhole Brain SPECT (#1329)

W. B. Huang1, G. S. Mok1, 2

1 University of Macau, Department of Electrical and Computer Engineering, Taipa, China, Macao Special Administrative Region
2 University of Macau, Centre for Cognitive and Brain Science, Institute of Collaborative Innovation, Taipa, China, Macao Special Administrative Region


Multi-pinhole (MPH) SPECT is recognized for improving diagnosis of Alzheimer’s disease (AD) and Parkinson’s disease (PD). Here we aim to investigate the relationship of pinhole numbers and projection views in MPH brain SPECT for a conventional dual-head SPECT scanner. The NCAT brain phantom was used to simulate 2 activity distributions, i.e., 99mTc-HMPAO for AD and 99mTc-TRODAT for PD. The target system spatial resolution was set to 12 mm for a 200 mm FOV to cover the patient’s head. We used a 3D analytical MPH projector to generate 64, 32, 16, 8, 4, 2 noise free and noisy projections for 1-15 pinholes respectively, which were designed based on our previous work (Chen et al 2017). The noise levels for different pinholes were modeled by scaling the clinical count level for 2 brain applications obtained from a LEHR parallel hole collimator. Noise free and noisy projections were reconstructed using the ML-EM algorithm up to data convergence. Normalized mean square error (NMSE) was calculated over the whole brain region while normalized standard deviation (NSD) was measured over a 3D uniform region. Based on the noise-free NMSE analysis for 64 to 2 projections, the optimum pinhole number increased from 3 to 11 for HMPAO and from 7 to 8 for TRODAT. Based on the NMSE-NSD analysis for noisy data, the optimum pinhole number increased from 4 to 10 for HMPAO and from 8 to 14 for TRODAT. We conclude that the optimum pinhole number increases along with the decrease of the projection numbers for both brain SPECT applications to provide more angular sampling. More pinhole number is needed in noise-free or low noise situation for a more diverse activity distribution, i.e., HMPAO in this study. An application with higher noise level, e.g., TRODAT-1 in this study, would be benefited from more numbers of pinhole.


This work was performed in part at SICC which is supported by SKL-IOTSC, Univ. of Macau. This work is supported by a FDCT Research Grant (0091/2019/A2).

Keywords: Multi-pinhole collimator, SPECT, brain, Alzheimer’s disease, Parkinson’s disease

NEMA NU 2-2018 Performance Evaluation of a New Generation 30cm AFOV DMI PET/CT (#360)

K. G. Zeimpekis1, F. A. Kotasidis2, M. Huellner1, A. Nemirovsky2, P. A. Kaufmann1, V. Treyer1

1 University Hospital Zurich, Nuclear Medicine, 8091, Zürich, Switzerland
2 GE Healthcare, 60661, Illinois, United States of America


The Discovery MI PET/CT is a modular SiPM based scanner with an axial FOV between 15-25cm depending on ring configuration (3, 4 or 5-rings). A new generation (Gen2) has been introduced and installed at our Nuclear Medicine Department at Zurich University Hospital, which includes amongst others, a reengineering detector module, featuring improved detector electronics and the ability to house an additional 6th ring, extending the FOV to 30cm axially. In this work we report on the performance of the 6-ring upgraded Gen2 system while values are also reported for the 5-ring configuration. Average sensitivity measured at 32.76 cps/kBq (~47% increase to 5-ring at 22.29cps/kBq) while NECR peaked at 434.3 kcps at 23.6 kBq/ml (~60% increase to Gen1 [3] and 39% to Gen2 5-ring). Contrast recovery ranged between 54.5-85.8% similar to 5-ring while the 6-ring provided lower background variability (2.3-8.5% for 5-ring vs 1.9-6.8% for 6-ring) and lower lung error (4.0% for the 5-ring and 3.16% for the 6-ring). Transverse/axial FWHM at 1cm (3.79/4.26mm) and 10cm (4.29/4.55), scatter fraction (40.2%), energy resolution (9.63%) and TOF resolution (389.6 psec at 0 kBq/ml) were in line to previously reported DMI values [3].

Keywords: PET/CT, NEMA, 6-Ring

Coupling of 18F-FDG and 18F-NaF PET/CT dynamic imaging for the detection of arterial inflammation (#1217)

A. Douhi1, M. S. Al-enezi1, A. Khalil2, M. Bentourkia1

1 University of Sherbrooke, Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada
2 University of Sherbrooke, Medicine, Sherbrooke, Québec, Canada


Arterial inflammation is an indicator of atheromatous plaque vulnerability to detach and to obstruct blood vessels in the brain thus causing vascular complications. To date there is no way to prevent the plaque detachment. In the present study, the metabolic activity of the cells in the plaque was assessed with 18F-FDG and the active microcalcification was assessed with 18F-NaF with PET/CT dynamic imaging in 5 volunteer aged 65-85 years. The artery segments were considered on each transaxial image slice on the CT and PET images. The arteries were considered calcified if they contained 3 or more adjacent pixels of density above 130 Hounsfield Units (HU). Regions of interest (ROIs) were drawn on the artery segments including blood by means of active contour technique on both CT and PET images. The arteries on PET images were corrected for partial volume effect and the radiotracer uptake was quantified with the Standard uptake value (SUV). A total of 418 arterial segments were analyzed, 263 were non-calcified and 155 had calcifications. The calcification in an artery segment was found as a single or multiple patterns. The statistical study showed a significant difference in 18F-FDG and 18F-NaF SUV values comparing non-calcified and calcified arteries. For 18F-NaF the mean SUV value was 0.46 ± 0.21 for non-calcified segments and 0.33 ± 0.15 for calcified segments, and similarly for 18F-FDG with 0.49 ± 0.12 for non-calcified and 0.36 ± 0.24 for calcified segments. The calcification intensity lowers the uptake of either 18F-NaF and 18F-FDG which might lead to concluding that the calcification is a sign of plaque stability, and the high uptake of 18F-FDG and 18F-NaF are an indicator of plaque vulnerability.

Keywords: Inflammation, Artery, PET/CT, 18F-FDG, 18F-NaF

MRI Compatibility Measurements of SIAT aPET (#110)

Z. Sang1, Z. Kuang1, J. Lee1, X. Wang1, N. Ren1, S. Wu1, D. Gao1, T. Zeng1, M. Niu1, L. Cong1, Z. Liu1, T. Sun1, Z. Hu1, Y. Li1, Y. Yang1

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


Dual-modality small animal PET/MRI imaging that provides multi-parameter temporal correlated information is a very powerful tool in biomedical research. A MRI compatible small animal PET insert called SIAT aPET was developed by using dual-ended readout depth encoding detectors to simultaneously achieve a uniform high spatial resolution and high sensitivity at Shenzhen Institutes of Advanced Technology (SIAT).  The PET insert has an out diameter 270 mm and could be inserted into a clinical MRI bore. An auxiliary small animal RF coil with transmitter and receiver was also developed and PET and MRI imaging could be performed simultaneously.  In this work, we investigated the mutual influence between the PET insert and a clinical 3T MRI system.  The results showed that the MRI imaging had negligible effect on the PET performance and the PET scanner had accepted effect on the MRI performance.  Simultaneous PET and MRI images of a Derenzo phantom and a mouse were also obtained.

Keywords: MRI compatible PET, small animal PET/MRI, Dual-end readout

A High Resolution TOF-PET prototype for Limited Angle Tomography (#407)

E. Lamprou1, G. Cañizares1, M. Vergara1, 2, H. Espinos1, D. Cascales1, L. F. Vidal1, J. Barrio1, C. Valladares1, M. Freire1, M. J. Rodriguez-Alvarez1, J. M. Benlloch1, F. Sanchez1, A. Gonzalez1

1 Institute for Instrumentation in Molecular Imaging (i3M), Valencia, Spain
2 Department of Imaging and Pathology, Division of Nuclear Medicine, KU Leuven, Leuven, Belgium


Nowadays, TOF information has become the main trend in PET instrumentation. TOF, besides offering an increased image Signal-to-Noise ratio, it also permits the development of new PET concepts, such as the Limited Angle Tomography (LAT). Without TOF, these concepts show a poor performance due to the poor sampling of polar angles.

In this work, we describe a dedicated TOF-PET scanner, designed for heart imaging of patients under stress condition. The system is called CardioPET and it is made out of 2-panels with an area of 17 ́ 11 cm2. It is based on 8×8 SiPM arrays, one- to-one coupled to LYSO crystals of 3 ́3 ́10 mm3. A total of 3072 resulted channels are read out using a commercially available ASIC-based DAQ. The first evaluation of the CardioPET shows that is capable of providing a system time resolution of 237 ps FWHM, with no degradation seen from the detector to the system level. We have carried out an in-depth performance evaluation. Some reconstructed images are also provided showing the performance of the system in terms of spatial resolution as well as the benefit of TOF in LAT applications.


This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No. 695536) and by the Spanish Ministerio de Economía, Industria y Competitividad under Grant TEC2016-79884-C2-1-R. The first author has also been supported by Generalitat Valenciana, Spain under grant agreement GRISOLIAP- 2018-026.

Keywords: Dedicated PET, LAT, SiPMs, TOF-PET, TOFPET2 ASIC

DOI calibration methods for the NeuroEXPLORER brain imager (#616)

A. Selfridge1, J. Schmall1, H. Li1

1 United Imaging Healthcare America, Houston, Texas, United States of America

On behalf of the NX Consortium


With both TOF and DOI capabilities, the NeuroEXPLORER requires additional calibration steps compared to many other clinical and research PET systems. Specifically, calibration of the continuous-DOI encoding detector is necessary to realize sub-250ps timing and to maintain a uniform spatial resolution for large acceptance angles and away from the center of the FOV. Based on the unique detector design and decoding method used in the NeuroEXPLORER, we have developed a method for data-driven calibration based on the count density within the flood or DOI histograms. The data driven DOI calibration workflow utilizes events acquired in a front-irradiation configuration, which is ultimately how data must be acquired in the completed system. We have validated the proposed method with coincidence and singles data acquired in both front- and side-on irradiation. A ground-truth data set using electronic collimation with a reference detector was first acquired. DOI thresholds based on side-irradiated singles and front-irradiated coincidences were then used to evaluate the accuracy of the data-driven DOI calibration methods. These values were then compared to GATE simulated front-irradiated coincidences for a detector with similar geometry. Data-driven DOI thresholds based on side irradiated singles had a mean error of 0.69 mm, while front irradiated coincidences had a mean error of 1.0 mm, which is sufficiently lower than the DOI resolution for the proposed detector.

Keywords: NeuroEXPLORER, NX, DOI, PET, Detector

Energy Spread Estimation of Radioactive Carbon Ion Beams Using Optical Imaging (#194)

H. G. Kang1, S. Yamamoto2, S. Takyu1, A. Mohammadi1, F. Nishikido1, T. Yamaya1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Department of Advanced Nuclear Medicine Science, Chiba, Japan
2 Nagoya University, Department of Radiological and Medical Laboratory Sciences in Nagoya University Graduate School of Medicine, Nagoya, Japan


Radioactive ion (RI) beams (e.g. 11C and 15O) combined with in-beam positron emission tomography (PET) allow treatment of cancer with exact range verification. However, one drawback of the RI beams (generated as secondary beams) over conventional stable heavy ion beams is the wide momentum distribution (i.e. energy spread) thereby causing the Bragg peak to shift as well as the distal-falloff length to widen. Hence, the energy spread of RI beams should be carefully measured during the quality assurance. In this study, we propose an optical imaging technique for the energy spread estimation of RI 11C ion beams. A PMMA phantom (10.0×10.0×9.9 cm3) was irradiated with a 11C beam (energy = 188.5 MeV/u, sigma = 4.5 MeV/u). Three different energy spreads of 1.1 MeV/u, 2.2 MeV/u and 4.0 MeV/u were obtained by using the momentum acceptances of 1%, 2% and 4%, respectively. An optical system consisting of a lens and a cooled CCD camera was used to acquire the optical images. The in-beam luminescence light images were obtained to extract the Bragg peak and distal falloff length information whereas the offline beam Cerenkov light images were obtained for the estimation of the stopping positions of the 11C ion beams. The energy spread of the 11C ion beam was estimated by two optical parameters: (1) distal-50% falloff length of the luminescence signals and (2) full-width at half maximum (FWHM) of the Cerenkov light in the beam direction. These parameters estimated the energy spread with the mean squared errors of 1.47×10-3 MeV/u and 9.8×10-5 MeV/u, respectively. In conclusion, the optical imaging can estimate the energy spread of RI 11C beams with high accuracy. The optical imaging (luminescence + Cerenkov) is a promising solution for the quality assurance of RI beams.

Keywords: Radioactive carbon ion beam, Optical imaging

Design and Initial Studies of a Four-layer DOI Detector for High Sensitivity and High Spatial Resolution Small Animal PET (#739)

W. He1, Y. Zhao1, X. Zhao1, J. Wang1, D. Prout2, A. Chatziioannou2, Q. Ren1, Z. Gu1

1 Shenzhen Bay Laboratory, Institute of Biomedical Engineering, Shenzhen, China
2 University of California,Los Angeles, Crump Institute for Molecular Imaging, Los Angeles, California, United States of America


A novel four-layer DOI detector comprised of LYSO and BGO scintillators is proposed for high sensitivity and high spatial resolution small animal PET imaging. Based on the proposed detector, a compact small animal PET system is designed to achieve an ultra-high sensitivity and a uniform sub-millimeter spatial resolution within the entire FOV. Preliminary experiments were performed to show the feasibility of the new DOI decoding methods. Furthermore, the thickness of four scintillator layers was optimized and system sensitivity was evaluated via Monte Carlo simulation.

The detector module is comprised of a stack of four alternating LYSO and BGO scintillator crystal arrays coupled to an 8×8 SiPM array. The crystal pitch is 1.0 mm for the top two layers, and 1.53 mm for the bottom two layers. The SiPM has a pixel pitch of 3.2 mm. The full size of the detector module is 25.8×25.8×26 mm3. The proposed small animal PET system has 4 detector rings, each comprised of 10 detector blocks, forming a bore diameter of 80 mm and an axial length of 103 mm.

Our previous work has shown that time-over-threshold (ToT) measured by the TOFPET2 ASIC can be used in combination with the integrated pulse charge to discriminate the LYSO and BGO layers. Preliminary experimental studies in this work show that convolutional neural network (CNN) can be used to model detector responses of different scintillator layers, and decode layers made of the same scintillators. Monte Carlo simulation shows that the quasi-optimal thickness for each layer from top to bottom was 5, 6, 7 and 8 mm under the criterion of keeping an approximately equal fraction of gamma events for each individual crystal element. Simulation shows that the preliminary small animal PET system had a peak sensitivity of 23.4% under 350-650 keV energy window.

In conclusion, preliminary results demonstrated both the feasibility of the DOI decoding method and the ultra-high sensitivity of the proposed PET system.

Keywords: small animal PET, DOI detector, high sensitivity and high spatial resolution

Small animals’ vital signs monitoring system for in vivo imaging applications (#870)

R. G. Oliveira1, P. M. M. Correia1, A. L. M. Silva1, F. M. Ribeiro1, P. M. C. C. Encarnação1, J. F. C. A. Veloso1

1 University of Aveiro, Institute for Nanostructures, Nanomodelling and Nanofabrication (i3N), Physics Department, University of Aveiro, 3810-193 Aveiro, Portugal, Aveiro, Portugal


The main goal of this work is the implementation of a setup of sensors capable of measuring the fundamental vital signs (heart rate, respiratory rate, body temperature, and oxygen saturation) in mice during imaging procedures. For this purpose, the measurement viability of the referred signals is evaluated for different sensors. This paper presents some tests with an optical module, piezoelectric sensor, thermistor, and electrocardiogram. The data acquired will be useful in the image reconstruction process by reducing the blurring caused by the mice motion, resultant from respiratory and cardiovascular movements. Additionally, this sensorization allows the control of animal’s wellbeing during imaging procedures. This work is being developed for integration of a monitoring bed into easyPET family systems, in particular into the iPET prototype, which is a high-performance preclinical PET scan.


This work was supported by A) project iPET CENTRO-01-0247-FEDER-039880, co-financed by the EU through FEDER; B) project i3N, UIDB/50025/2020 & UIDP/50025/2020, financed by national funds through the FCT/MEC; C) project PTDC/EMD-EMD/21402020; and D) grants to R.G. Oliveira (SFRH/BI/10638/2020), F.M.Ribeiro (SFRH/BD/137800/2018) and P.M.M.C. Encarnação (SFRH/BD/143964/2019) through the FCT, Potugal.  

Keywords: body temperature, heart rate, image gating, respiratory rate, vital signs monitoring

Compton-Enhanced PET Imaging with High-Resolution 3-D CZT Imaging Detectors (#1030)

Y. Jin1, P. Tanton1, M. Streicher3, H. Yang3, S. Brown3, Z. He3, 4, L. - J. Meng1, 2

1 University of Illinois at Urbana and Champaign, Department of Nuclear, Plasma, and Radiological Engineering, Urbana, Illinois, United States of America
2 University of Illinois at Urbana and Champaign, Department of Bioengineering, Urbana, Illinois, United States of America
3 H3D, Inc., Ann Arbor, Michigan, United States of America
4 University of Michigan at Ann Arbor, Department of Nuclear Engineering and Radiological Sciences, Ann Arbor, Michigan, United States of America


In this paper, we report the experimental results of  positron emission tomography (PET) imaging with Compton kinemetics enhancement that employs large volume 3-D position sensitive Cadmium Zinc Telluride (CZT) detectors. The large volume CZT detectors used in this system offer an high intrinsic spatial resolution of approximately 0.5mm full width at half maximum (FWHM) in all three dimensions and an ultrahigh energy resolution (3 keV @ 200 keV, 4.5 keV @450 keV, 5.4 keV @511 keV) [1]. Furthermore, the detector has the ability to precisely detect multiple interactions induced by a single incident gamma ray, and therefore allow the use of Compton kinematics to reject chance coincidence events and image the distribution of the object as Compton camera.

We have carried out a series of experiments using a prototype PET system with 4 square panels of 3-D CZT detectors to study the influence of depth-of-interaction (DOI) resolution and the benefits of Compton capabilities on the resultant imaging performance.

In this presentation, we will demonstrate the experimental performance of rejection of chance coincidence events based on Compton kinematics, its impact on PET image quality,  and combined PET and near-field Compton reconstruction.


We would like to thank NIBIB (EB004940) and NHLBI (R01HL145786) for their partial support for this work.

Keywords: ultrahigh resolution positron emission tomography (PET), near-field Compton responses, 3-D position-sensitive cadmium zinc telluride (CZT) detector

Design Study of a Brain SPECT System based onSynthetic Compound-Eye Camera Design and Micro-Ring Apertures for Ultrahigh Resolution Brain Imaging (#1168)

E. M. Zannoni1, 2, C. Yan2, L. J. Meng2, 1

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


Brain-dedicated SPECT imaging systems have the stringent requirement to simultaneously provide high spatial resolution and high sensitivity, due to the microscopic anatomy under evaluation (~3-4 mm) and the low radiotracer uptakes (~3.5-7.0%). However, clinical SPECT instrumentation is still mainly based on cumbersome scintillation detectors that are characterized by poor energy and spatial resolution.

In this presentation, we will discuss a conceptual design study of a high-performance brain SPECT system, a compact and stationary SPECT system based on pixelated Cadmium Zinc Tellurium (CZT) detectors, and a Synthetic Compound-Eye (SCE) gamma camera design equipped with micro-rings. The preliminary design presents 502 high-performance CZT detectors, 2×2×0.5 cm3 in size and offering 80 × 80 pixels of 250 μm ×250 μm pitch. The simulated detector presents the same form factor and properties of the CTZ/HEXITEC ASIC detector, whose excellent energy resolution has been experimentally demonstrated (1.06±0.2 keV at 140 keV). Each detector module is then coupled with a micro-ring aperture, characterized by a narrow annulus width that guarantees an ultra-high spatial resolution, and an external diameter that increases the overall aperture open fraction. From Monte Carlo simulations, the combination of high-performance solid-state sensors and non-conventional apertures provides a peak geometrical sensitivity of 1.38% while preserving a remarkable spatial resolution of <4 mm.


We would like to acknowledge the National Institute of Biomedical Imaging and Bioengineering (NIBIB) as funding agency.

Keywords: brain SPECT imaging, ultra-high resolution, CZT detectors, micro-ring collimator

Single and Random Rate Analysis of A Conformal Prism-PET Brain Scanner (#1243)

Z. Wang1, E. Petersen1, X. Zeng2, X. Cao2, W. Zhao3, A. Goldan3

1 Stony Brook University, Department of Biomedical Engineering, Stony Brook, New York, United States of America
2 Stony Brook University, Department of Electrical and Computer Engineering, Stony Brook, New York, United States of America
3 Stony Brook University, Department of Radiology, Stony Brook, New York, United States of America


It has been argued that the brain-dedicated PET scanner with compact geometry suffers from increased dead- time and random coincidences due to a higher single rate. In this work, we analytically model the single efficiency of our proposed brain-dedicated Prism-PET scanner and compared it to the state-of-art Biograph Vision (BV) using the GATE Monte-Carlo simulation toolbox. Our results suggest that the Prism-PET brain scanner receives fewer singles from the phantom that is outside of axial field-of-view (FOV). For a 700 mm length phantom placed at the edge of axial FOV, Prism-PET only has a 10% higher single rate than BV. We believe the reason that Prism-PET has a higher random rate is because it has a larger FOV-to-geometry volume ratio, which may include a larger percent of randoms in the FOV during the count rate evaluation. By applying a smaller coincidence window, the Prism-PET brain scanner can reach a peak NECR of 649 kcps at 35.75 KBq/cc, enabling high sensitivity quantitative brain imaging applications.

Keywords: Prism-PET brain scanner, Single efficiency, Random coincidence rate

Performance Assessment of a High-Resolution Small Animal CZT PET System (#1449)

A. Groll1, R. Stanford-Hill2, C. S. Levin1

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


This work focuses on the study of the performance of a large volume small animal CZT PET system sub-assembly. Currently, two full panels of an anticipated four-panel system have been completed with a total of 24 dual-CZT crystal modules (48 crystals). Each dual-module in this work uses two monolithic CZT crystals, each with dimensions of 40 x 40 x 5 mm3. The dual crystal CZT Module is arranged in an anode-cathode-cathode-anode configuration with 39 anode strips and 8 cathode strips per crystal, a total of 38 steering electrodes reside between the anode strips. A total of 1872 anode strips and 384 cathode strips are supported by custom readout ASICs. The energy resolution was optimized to 10% for panel one and 7% for panel two for 511 keV with the full system achieving 8.1% FWHM. Coincidence timing resolution was optimized to 34 ns using an iterative convex optimization process to remove channel-to-channel delays.


This work is supported by NIH-NCI T32 CA118681 and Stanford Bio-X. We thank Harrison Lin and Prof. Joseph DeSimone for providing rapid prototyping support via Carbon3D printers.

Keywords: PET, High Resolution, CZT, Dual Panel, Preclinical

Tomographic Images given by Small Number X-ray Transmission Measurements with using Material Thickness Distributions (#121)

I. Kanno1, D. Ito1

1 Kyoto University, Nuclear Engineering, Kyoto, Japan


In an X-ray computed tomography (CT) diagnosis with iodine contrast agent, nearly 1,000 X-ray transmission measurements are carried out for reconstructing a CT image. As a result, a CT measurement has high dose exposure. If an iodine contrast agent distribution is obtained with small number of transmission measurements even though the distribution image is relatively rough, a cancer screening check is performed with low dose exposure.

We have developed a two-dimensional “transXend” detector which measures X-rays as electric current and gives incident X-ray energy spectrum by analysis. With previously estimated look-up tables for measured electric currents, thickness distributions of materials in a phantom are obtained for acrylic, iodine and aluminum. These thickness distributions are measured from twelve directions. With choosing two-, four- and twelve-thickness distributions, tomographic images for acrylic, iodine and aluminum are given. The tomographic image with four-direction measurements shows excellent iodine distribution, although the one with two-direction measurements results in an ambiguous image.

AcknowledgmentThe authors are thankful to Dr. K. Shimomura at Kyoto College of Medical Science for his fruitful discussions.
Keywords: computed tomography, low dose exposure, X-ray, contrast agent, material thickness distribution

Multi-Material Decomposition Methods of Dual-Energy CT Images, Revisited (#372)

D. Shi1

1 Liaoning Campo Medical Imaging Tech. Ltd. Co., Benxi, China


In this work, we propose first a two-stage algorithm for multi-material decomposition (MMD) of ducal-CT images, which is equivalent to the existing method in [1]. The novelty of our proposal is that in the first stage we employ a simple left-on test algorithm to classify a pixel of the dual-CT image into two types, namely unsolvable and solvable from a geometric point of view. In the second stage we decompose those unsolvable and solvable pixels into basis materials by the projection method and the material list method in [1], respectively. In the existing method in [1], it is assumed that there are at most three basis material components in each pixel and various basis material components among pixels. Our second proposed algorithm relax this assumption by removal of the limitation that at most three material components can exist in each pixel, i.e., our proposed algorithm assumes that any number (more than three) of basis material components can exist in each pixel to be considered. We achieve this goal by using a constrained minimization scheme. A patient data set was employed to validate our proposed algorithm.

Keywords: Dual-Energy CT, Multi-Material Decomposition

Spectral distortion correction of photon-counting CT with machine learning (#552)

K. Murata1, K. Ogawa2

1 National Astronomical Observatory of Japan, Tokyo, Japan
2 Hosei University, Faculty of Science and Engineering, Tokyo, Japan


We propose a spectral distortion correction method for a photon-counting CT system with machine learning. Although the most important advantage of the photon-counting CT system is material-decomposition capability, it is very sensitive to the most serious problem, the pulse pile-up effect. This leads to spectral distortion and material densities could not be accurately measured. In this study, we constructed a neural network for spectral distortion correction. The performance of our network was investigated with a simulation. Our network trained with a data set of 1000 pairs of original and distorted spectra. The data set was prepared with X-ray spectra whose intensity was high enough to significantly distort the spectra, which were attenuated by targets with various lengths to produce many spectral shapes. We also investigated the loss-function dependence of the spectral distortion correction. We found that our network remarkably corrected the spectral distortion and material densities could be precisely measured. We also found that MAE loss function leads to more accurate measurements. These results suggest that our network is effective for material decomposition and optimizing loss function could lead to better results.

Keywords: medical imaging, computed tomography, photon counting, pulse pileup

Improvement of the spatial resolution with a deconvolution method for a multi-pinhole SPECT system (#772)

M. Okoshi1, K. Murata2, K. Ogawa3

1 Hosei University, Graduate School of Science and Engineering, Tokyo, Japan
2 National Astronomical Observatory of Japan, Tokyo, Japan
3 Hosei University, Faculty of Science and Engineering, Tokyo, Japan


We propose a method to improve the spatial resolution of an image for a static multi-pinhole single photon emission CT (SPECT) system. This SPECT system has a great advantage of measuring dynamic function of organs, which could not be performed by conventional systems. However, the image quality is very sensitive to the pinhole size of a collimator. In the case of a large pinhole, the widely applied 7-rays method could not sufficiently correct the image resolution. Hence, we propose a point spread function (PSF) deconvolution method for a multi-pinhole SPECT system. This method assumes a shift-invariant PSF derived from the pinhole geometry, and performs PSF deconvolution on projection data. The feasibility of this method was investigated with a simulation. We assumed a static SPECT system with a multi-pinhole collimator, and projection images were obtained by a Monte Carlo method. The reconstructed images with our method had significantly higher spatial resolution and reproduced more detail structures compared with those of the 7-rays method. The images also showed higher PSNRs than the conventional ones. These results indicated the effectiveness of our proposed method.

Keywords: static SPECT system, multi-pinhole collimator, deconvolution method, point spread function

Evaluation of Time-of-Flight Scatter Rejection in X-Ray Radiography (#1180)

J. Rossignol1, 2, ?. Pilon1, G. Bélanger1, 2, Y. Bérubé-Lauzière2, R. Fontaine1, 2

1 Université de Sherbrooke, Interdisciplinary institute of technological innovation 3IT, Sherbrooke, Québec, Canada
2 Université de Sherbrooke, Departement of electrical engineering and computer engineering, Sherbrooke, Québec, Canada


Previous simulation studies have shown the ability of time-of-flight (TOF) measurements of X photons to discriminate between scattered and ballistic photons. In an initial study, this has shown its potential to significantly reduce the scatter contribution in X-ray computed CT, especially in cone-beam CT, leading to an improvement in image quality or to a dose reduction. In X-ray radiography, scatter contribution also causes a degradation of the contrast-to-noise ratio (CNR). This study, herein, investigates the possibility of using the TOF scatter rejection method, developed for CT, in X-ray radiography. The radiography of a head and a torso is simulated using GATE. Images were generated with no scattering, no scatter rejection and with TOF scatter rejection in systems with a total timing jitter of 10, 100 and 200 ps. When compared to an image obtained with no scatter rejection, a 200 ps timing jitter removes half of scattered photons and the CNR is improved from 51% to 69% of the no scattering CNR. At 10 ps, 95% of scattered photons are removed and the CNR is restored to 96% of the no scattering CNR.

Keywords: Time-of-flight, X-ray radiography, scatter rejection, scatter noise, medical imaging

Low-dose Direct PET Image Reconstruction Using Channel Attention for Deep Neural Network (#1)

T. Yin1, T. Obi2

1 Tokyo Institute of Technology, Department of Information and Communications Engineering, Yokohama, Japan
2 Tokyo Institute of Technology, Institute of Innovative Research, Yokohama, Japan


Positron emission tomography (PET) is a medical imaging approach widely used in various clinical applications. There is significant value in low-dose PET image reconstruction, because radiation risk is reduced when patients are injected with lower dose of radiotracer. However, this results in a high level of noise in emission data, which degrades the quality of activity distribution images. In this paper, we propose a deep neural network for low-dose PET reconstruction. Using time-of-flight (TOF) sinograms as inputs, it generates high-quality quantitative PET images directly. Specifically, we utilize an encoder-decoder to transfer projections in sinogram domain to activity maps in image domain. Then the outputs of previous stage are restored using a deep neural network with channel attention modules. Residual connections allow abundant low-level features to be bypassed, while channel attention blocks (CABs) capture high-level features by extracting channel statistics. A channel attention mechanism learns nonlinear cross-channel interactions among channels and generates high-resolution feature maps. We inject supervision to both the initial output after domain transformation and the final output. The loss function is comprised of the mean square error (MSE) of two outputs and their edge losses. Experimental results show that our method results in a normalized root MSE (NRMSE) of 1.579%, a structural similarity index measure (SSIM) of 0.631 and a peak signal-to-noise ratio (PSNR) of 16.66 dB, outperforming maximum likelihood attenuation correction factors (MLACF) algorithm following nonlocal mean (NLM) post processing method, DeepPET, and a single encoder-decoder in the earlier stage of our network in quantitative evaluations. The qualitative analysis demonstrates that the proposed approach is capable of preserving fine details. This method shows promise in improving PET image quality with low-dose emission data.

Keywords: PET, Image reconstruction, Deep learning, Channel attention

Multi-scale Wavelet Kernel-based PET Reconstruction Using MR Side Information (#174)

Z. Ashouri1, 2, G. Wang3, R. Dansereau1, R. deKemp2

1 Carleton university, Ottawa, Ontario, Canada
2 Ottawa Heart institute, Ottawa, Ontario, Canada
3 UC Davis Health, Radiology Department, Davis, California, United States of America


Positron emission tomography (PET) image reconstruction is challenging due to low number of detected coincidence events. In PET/MR, the MR image prior information can be used to improve PET image quality, for example through kernel-based image reconstruction. Typically, kernel methods use a Gaussian kernel; however, a Gaussian kernel tends to over-smooth details in the reconstructed image. To better preserve edges, the wavelet kernel has been introduced, and in this work, we extend the wavelet kernel method from single-scale to multi-scale. Multi-scale wavelets are expected to perform better at extracting data features. Wavelets are used and the results compared with the Gaussian kernel and maximum likelihood expectation maximization (MLEM) for PET image reconstruction. Our results suggests that the multi-scale wavelet kernel produces reconstructed images with better quality than the single-scale wavelet, Gaussian kernel and MLEM.

Keywords: PET image reconstruction, wavelet kernel method, magnetic resonance prior, multi-resolution wavelet

Implementation and image quality benefit of a hybrid-space PET point spread function (#314)

T. W. Deller1, S. Ahn2, F. P. Jansen1, G. Schramm3, K. A. Wangerin1, M. G. Spangler-Bickell1, C. W. Stearns1, M. M. Khalighi4

1 GE Healthcare, Waukesha, Wisconsin, United States of America
2 GE Global Research, Niskayuna, New York, United States of America
3 KU Leuven, Department of Imaging and Pathology, Division of Nuclear Medicine, Leuven, Belgium
4 Stanford University, Department of Radiology, Stanford, California, United States of America


Detector response modeling during reconstruction of PET data improves image resolution. Such modeling is typically performed either in sinogram (projection) domain, or in image domain. This work proposes a hybrid approach, with a portion of the point spread function (PSF) modeling in each domain. The approach includes a spatially varying radial smoothing of the sinogram data (or broadened line of response if performing list mode reconstruction), combined with a spatially invariant image smoothing. The image-based smoothing component mitigates high-frequency artifacts that can emerge from the projectors when small pixel sizes are used with low levels of post-smoothing; this benefit is important for many high-resolution PET imaging approaches. Unlike fully-image-based PSF approaches, this method integrates well with motion-corrected list-mode reconstruction by applying the appropriate PSF kernel for each LOR, even when its position is adjusted for motion. The hybrid approach opens the way to isotope-dependent positron range imaging, by allowing the use of different image-based kernels for isotopes with different positron range, while preserving the same projection-based kernel (which is a function of detector design only). The effectiveness of the method is demonstrated with a brain 18F-FDG dataset and spatial resolution point sources.

Keywords: High-resolution imaging, PET, point spread function, spatial resolution

FDG-guided image reconstruction for dual-tracer PET (#447)

H. Wang1, 2, X. Zhang3, W. Fan3, C. Zhou3, J. Yuan4, Q. He4, Y. Yang1, D. Liang1, Z. Hu1

1 Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Lauterbur Research Center for Biomedical Imaging, Shenzhen, China
2 University of Chinese Academy of Sciences, Shenzhen College of Advanced Technology, Beijing, China
3 Sun Yat-sen University Cancer Center, Department of Nuclear Medicine, Guangzhou, China
4 Shanghai United Imaging Healthcare, Central Research Institute, Shanghai, China


Clinically, single radiotracer positron emission tomography (PET) imaging is a commonly used examination method; however, since each radioactive tracer can reflect the information of only one kind of cell, it can easily cause false negatives or false positives during disease diagnosis. Therefore, reasonably combining two or more radiotracers is recommended to improve the accuracy of diagnosis and the sensitivity and specificity of the disease when conditions permit. This paper proposes incorporating 18F-fluorodeoxyglucose (FDG) as a higher-quality PET image to guide the reconstruction of other lower-count PET datasets to compensate for the lower image quality by using a kernel algorithm. In a simulation study, the proposed method outperformed other algorithms by at least 3.11% in signal-to-noise ratio (SNR) and 0.68% in contrast recovery coefficient (CRC), and it reduced the mean absolute error (MAE) by 8.07%. The improvement was also verified in a patient data case study; in particular, the proposed method gained a 28.3% higher SNR than the other algorithms. The proposed FDG-guided KEM algorithm can effectively utilize and compensate for the tissue metabolism information obtained from dual-tracer PET to maximize the advantages of PET imaging.


This work was supported by the National Natural Science Foundation of China (32022042, 81871441).

Keywords: dual-tracer positron emission tomography (PET), FDG-guided, image reconstruction

Positron Range Modeling for Low-Dose Rb-82 Cardiac PET (#562)

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

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


The resolution degradation in PET is usually compensated by incorporating point spread function (PSF) kernels in the reconstruction system matrix. These PSF kernels can be either simulated using Monte Carlo simulations or measured using F-18 point sources in a real scanner. High energy emitters including Rb-82, however, have long positron ranges which can easily exceed ~5mm FWHM depending on the medium, and therefore causing relatively poor image resolution when using kernels measured by F-18 point sources. The positron range of Rb-82 is conventionally modeled as a local, spatially invariant kernel generated by analytical models or Monte Carlo simulations. However, these approaches do not include the spatially varying nature of resolution degradation in a real PET system. In this study, we propose to decompose the PSF kernel into two components, one based on spatially variant F-18 point source measurements taken at multiple locations in a real scanner and a second component that is a spatially invariant differential kernel that captures the differences between the analytical F-18 and Rb-82 kernels. The resulting kernel is approximated by a parameterized 3D asymmetric Gaussian function. We compared the proposed kernels with the measured F-18 point source kernels on five rest/stress myocardial perfusion studies acquired with either 25mCi or 30mCi Rb-82 infusion on a SiPM-based PET/CT scanner. The results show that incorporating the positron range in the PSF kernels improved the image resolution and myocardium-to-blood pool contrast with lower background noise.

Keywords: Positron range modeling, PSF reconstruction, Cardiac PET

Unsupervised training for PET sinogram denoising using Poisson unbiased risk estimator (#699)

H. Kim1, D. Hwang2, J. S. Lee2, 3, S. Y. Chun4

1 UNIST, Department of Electrical Engineering, Ulsan, Republic of Korea
2 Seoul National University, Department of Biomedical Sciences, Department of Nuclear Medicine, Seoul, Republic of Korea
3 Seoul National University, Institute of Radiation Medicine, Medical Research Center, Seoul, Republic of Korea
4 Seoul National University, Department of Electrical and Computer Engineering, Seoul, Republic of Korea


Recently, there have been many studies including deep learning-based approaches for enhancing PET image quality by reducing high noise. However, most of state-of-the are deep learning based denoising methods are supervised training that requires clean ground truth images. Unfortunately, obtaining noiseless PET images are often infeasible, thus unsupervised training scheme for PET sinogram denoising is desirable. We investigated the feasibility of unsupervised training for PET sinogram denoising using Poisson unbiased risk estimator (PURE) to achieve relatively high signal to noise ratio (SNR) without clean ground truth images. We trained a deep neural network based denoiser using our PURE loss with low count data (60 seconds from list-mode) only and then evaluated the result with full count (100 seconds). Our proposed method improved the reconstruction quality on low count measurements by 21.7% in root mean squared error.


This work was supported in part by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education, South Korea (NRF-2017R1D1A1B05035810) and in part by the Interdisciplinary Research Initiatives Programs by College of Engineering and College of Medicine, Seoul National University(2021).

Keywords: PET, denoising, unsupervised training

The Reconstruction Method Using Compressed Sensing and Convolutional Neural Network for PROPELLER MRI in Head (#855)

Y. Matsumoto1, K. Hori1, K. Tadano1, S. Kuhara1, Y. Endo1, T. Hashimoto1

1 Kyorin University Graduate School of Health Sciences, Department of Radiological Science, Mitaka, Japan


PROPELLER (Periodically Rotated Overlapping ParallEL Lines with Enhanced Reconstruction) MRI (Magnetic Resonance Imaging) is a method of reconstructing from the collected data of rectangular regions (blades) rotated around the origin of the k-space. This method can compensate for the motion of the subject by using the phase and rotation between the blades. However, it has the disadvantage of increasing the imaging time when many blades are collected for the motion correction. In this study, we attempted to improve the image quality for PROPELLER MRI in head by reconstructing from less data with compressed sensing (CS) and convolutional neural network (CNN), which is one of the deep learning methods. We compared three patterns of the reconstruction method: A) with only CS, B) with only CNN, and C) with both CS and CNN. In the method with CNN, the network that could predict artifacts included in input image was used and an artifact reduced image could be generated by subtracting the predicted artifacts from input image. The T2-weighted head images were used for learning of CNN. The possibility of reconstruction methods was investigated in the various sampling rate that was simulated by changing the blade width and the number of blades. When the sampling rate was less than 35%, the method with CS and CNN gave the best evaluation value, therefore, it was suggested that the image quality could be improved by reconstructing with CS and CNN when the sampling rate was low.


The IXI Dataset was used as the training image for the network in this study.

Keywords: Compressed sensing, Convolutional neural network, Image reconstruction, Magnetic resonance imaging, PROPELLER

Dual-Energy CT Reconstruction with Convolutional Analysis Operator Learning (#983)

S. L. Alfonso Garcia1, A. Perelli1, A. Bousse1, D. Visvikis1

1 Université de Bretagne Occidentale (UBO), LaTIM, INSERM, Brest, France


Dual Energy Computed Tomography (DECT) has the potential to improve contrast and reduce artifacts as compared to monoenergetic computed tomography (CT). In this paper, we propose an efficient method to exploit common spatial features within attenuation images at different energies. We develop an optimization method that jointly reconstructs the attenuation images at low and high energies with a mixed norm data-driven regularization. In particular, the regularization term promotes the joint sparsity between features that are obtained by pre-trained filters through the convolutional analysis operator
learning (CAOL) algorithm. Numerical experiments on sparse views DECT reconstruction show that the proposed joint method outperforms both CAOL applied on each energy independently and the model-based iterative reconstruction (MBIR) method using edge-preserving regularization.

Keywords: Dual-energy Computed Tomography, Convolutional Learning, Model-based Image Reconstruction

MLEM-based Reconstruction with a New Stochastic Enhancement for Compton Imaging (#1178)

M. - E. Tomazinaki1, M. Mikeli1, E. Stiliaris1, 2

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


Considering the lack of algorithmic schemes to depict objects of interest in Compton imaging in a realistic and efficient way, a new List Mode-MLEM-based approach is presented in this work. The developed algorithm is based on the standard LM-MLEM enhanced with a new probabilistic formula of the well-known system matrix based on the angular information of the incident photons. An accumulated probability, involving the Compton scattering angle, allows all detected events to originate from any pixel of the image giving a different weight. Hence, the proposed technique is free of any other conical section calculations normally introduced in a geometrical approach or any complicated mathematical expressions based on deterministic models. For a given focal plane, any point of the image to be reconstructed constitutes a potential solution assigned to a probability, which can be estimated for all events. Based on maximizing likelihood principles, the proposed model calculates the difference of the scatterer-to-source angle (θ0), as it is determined by the deposited energy on the absorber, and any potential scattering angle (θJ), specified by the position coordinates on the reconstruction matrix. The variation of the angular acceptance and the overall efficiency of the system for different source depths are used to perform volume corrections and further image optimization. Quantitative results of the proposed method tested with a plethora of source geometries for a simple two-stage scatterer-absorber Compton prototype simulated in the GEANT4/GATE environment are presented and discussed.


This research is co-financed by Greece and the European Union (European Social Fund -- ESF) through the Operational Programme ''Human Resources Development, Education and Lifelong Learning 2014--2020'' in the context of the project ''Reconstruction of Tomographic Image in Compton Camera System'' (MIS 5048146).

Keywords: Compton Camera, Iterative Reconstruction Algorithms, GEANT4/GATE, LM-MLEM

Towards Accurate Partial Volume Correction – Perturbation for SPECT Resolution Estimation (#37)

R. Gillen1, 2, K. Erlandsson1, A. M. Denis-Bacelar2, K. Thielemans1, B. F. Hutton1, S. McQuaid1

1 University College London, Institute of Nuclear Medicine, London, United Kingdom
2 National Physical Laboratory, Medical, Marine and Nuclear Department, Teddington, United Kingdom


The accuracy of quantitative SPECT imaging is limited by relatively poor spatial resolution - known as the Partial Volume Effect. There is currently no consensus on the optimal Partial Volume Correction (PVC) algorithm in the application of SPECT oncology imaging. However, several promising candidates require information on the reconstructed resolution - usually in the form of the point spread function (PSF). A particular challenge that SPECT poses for PVC is that the resolution is known to vary with position in the field-of-view, as well as activity distribution and reconstruction method. In this work, we assessed the viability of using perturbation to measure case-specific resolution. A small point source was used to measure the resolution in phantoms designed to replicate the issues encountered in oncology imaging. This included anthropomorphic phantoms which had not previously been examined in perturbation applications. Results demonstrate that, provided that a sufficient number of iterations is used for the image reconstruction, perturbation can be used to measure a case-specific PSF and can be used to improve the accuracy of PVC in oncology SPECT imaging.

Keywords: SPECT, quantification, partial volume correction, resolution, perturbation

Attenuation Correction for Myocardial Perfusion SPECT using Multi-stream 3D Residual Network (#486)

C. - Y. Yu1, K. - H. Lue2, H. - T. Yang3, T. - P. Su3, H. - H. Lin1, 3

1 Chang Gung University, The Institute for Radiological Research, Taoyuan, Taiwan
2 Tzu Chi University of Science and Technology, Department of Medical Imaging and Radiological Sciences, Hualien, Taiwan
3 Chang Gung Memorial Hospital Keelung Branch, Department of Nuclear Medicine, Keelung, Taiwan


Myocardial perfusion imaging (MPI) with single-photon emission computed tomography (SPECT) is one of the major imaging modalities for patients with potential coronary artery disease (CAD). However, SPECT MPI is susceptible to soft tissue attenuation. We conducted a prospective study to develop a novel approach for attenuation correction using deep learning (DL) algorithm and SPECT projection datasets. In the deep neural network architecture, two streams presenting two data inputs including projection data of photopeak and scatter window were used, while the target was the CT-based attenuation correction (CTAC) projection data. Myocardial perfusion SPECT/CT images of 100 patients were included for training and testing. Results showed that SPECT images corrected by the DLAC method were highly consistent with the images using the CTAC method, while Chang’s method caused an overcorrection of counts, especially at the tissue surround the heart. In comparison to SPECT images using CTAC, the normalized mean square error (NMSE) was 0.03 ± 0.04 for DLAC and 0.11 ± 0.06 for Chang’s AC, whereas structural similarity (SSIM) was 0.98 ± 0.03 for DLAC and 0.91 ± 0.05 for Chang’ AC. In conclusion, we established a new DL-based SPECT MPI AC method with similar image quality as CTAC which allows AC in the stand-alone SPECT system and decreases the radiation burden of CT.

AcknowledgmentThe work was financed by Ministry of Science and Technology of Taiwan (Project No. 108-2314-B-182-029 -MY2) and the Chang Gung Memorial Hospital (Project No. CMRPD1K0441).
Keywords: Attenuation correction (AC), Single-Photon Emission Computed Tomography (SPECT), Deep learning (DL)

Deep Active Learning Model for Adaptive PET Attenuation and Scatter Correction in Multi-Centric Studies (#744)

I. Shiri1, A. Sanaat1, E. Jafari2, R. Samimi3, M. Khateri4, P. Sheikhzadeh5, P. Geramifar6, H. Dadgar7, H. Arabi1, M. Assadi2, C. Uribe8, A. Rahmim9, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Genève, Switzerland
2 Bushehr Medical University Hospital, The Persian Gulf Nuclear Medicine Research Center, Department of Molecular Imaging and Radionuclide Therapy, Bushehr, Iran (Islamic Republic of)
3 Shahid Beheshti University, Department of Medical Radiation Engineering, Tehran, Iran (Islamic Republic of)
4 Islamic Azad University, Department of Medical Radiation Engineering, Science and Research Branch,, Tehran, Iran (Islamic Republic of)
5 Tehran University of Medical Sciences, Department of Nuclear Medicine, Imam Khomeini Hospital Complex, Tehran, Iran (Islamic Republic of)
6 Tehran University of Medical Sciences, Research Center for Nuclear Medicine, Shariati Hospital, Tehran, Iran (Islamic Republic of)
7 Imam Reza International University, Cancer Research Center, Razavi Hospital, Mashhad, Iran (Islamic Republic of)
8 University of British Columbia, Department of Radiology, Vancouver, British Columbia, Canada
9 University of British Columbia, Department of Radiology, Vancouver, British Columbia, Canada


Clinical centers are equipped with different scanners and use different acquisition/reconstruction protocols, and these may vary over time in a given center. As new scanners and protocols emerge, deep learning model performance may deteriorate significantly, even for models built using large datasets, hence the need to update the models. The aim of the current study was to apply a deep active learning model for adaptive PET attenuation and scatter correction in multi-centric studies. We enrolled 1110 18F-FDG and 950 68Ga-PSMA PET/CT images from 3 and 4 different centers, respectively. We implemented a deep residual network architecture with 20 blocks for different-level feature extractions. First, the deep neural network was trained on 18F-FDG PET images. Since the radiotracer distribution is different from 68Ga-PSMA PET, we used body fine tuning transfer learning to transfer ASC knowledge between radiotracers. We build deep learning-based ASC models on 68Ga-PSMA and then applied active learning approaches to build center-specific ASC model. We trained a 2D deep neural network for direct generation of CT-based attenuation corrected PET images from non-attenuation-scatter corrected (NAC) images of 68Ga-PSMA patients. For model evaluation, voxel-wise mean error (ME), relative error (RE%), absolute relative error (ARE%) and structural similarity index (SSIM) were calculated between ground truth CT-based attenuation/scatter corrected and predicted PET images using the deep learning algorithm. We achieved a ME of 0.22±0.05, RE of 2.72±7.5%, ARE of 10.0±4.5% and SSIM of 0. 98±0.02 in the test set. Overall, we applied transfer learning to transfer knowledge between different PET radiotracers and built specific models for each center separately using active learning approaches.

AcknowledgmentThis work was supported by the Swiss National Science Foundation under grant SNRF 320030_176052; the Swiss Cancer Research Foundation under Grant KFS-3855-02-2016
Keywords: Active Learning, Deep Learning, PET, Multicenter

Deep-PVC: A Deep Learning Model for Synthesizing Full-Dose Partial Volume Corrected PET Images from Low-Dose Images (#845)

A. Sanaat1, A. S. Boehringer1, A. Ghavabesh2, I. Shiri1, Y. Salimi1, H. Arabi1, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Genève, Switzerland
2 Shahid Chamran University of Ahvaz, Electrical Engineering Department, Ahvaz, Iran (Islamic Republic of)


Moderating the injected activity and/or reducing acquisition time to minimize potential radiation hazards and increase patient’s comfort are important trends in PET. This work aims to assess the performance of regular full-dose partial volume corrected (FD-PVC) image synthesis from fast/low-dose (LD) brain PET images using deep learning techniques for 18F-FDG, 18F-Flortaucipir, and 18F-Flutemetamol radiotracers. PVC techniques are designed to correct the spillover effect caused by the poor spatial resolution of PET scanners. The effect of resolution can be modeled as a convolution of the true image with a three-dimensional point-spread function (PSF). Clinical brain PET/CT studies of 100 patients for each radiotracer were prospectively employed for LD to FD-PVC PET image conversion. 5% of the events were randomly selected from the FD list-mode PET data to simulate a realistic LD acquisition. A modified cycle-consistent generative adversarial network (CycleGAN) model was implemented to predict FD-PVC PET images. Quantitative analysis using established metrics, including the peak signal-to-noise ratio (PSNR), structural similarity index metric (SSIM), root mean square error (RMSE), and standardized uptake value (SUV) bias, were performed. The R2 values were 0.95, 0.90, and 0.93 for 18F-FDG, 18F-Flortaucipir, and 18F-Flutemetamol, respectively. PSNR and SSIM values of 33.62±2.35 and 31.81±3.7, 34.44±4.11 and 0.96±0.02, 0.96±0.03, and 0.95±0.04 were obtained for 18F-FDG, 18F-Flortaucipir, and 18F-Flutemetamol, respectively. Our model provided image quality similar to PVC images from low dose PET images through a CycleGAN model.

Keywords: brain imaging, low-dose imaging, deep learning, PVC, Partial Volume Correction

Transmission-less attenuation correction for full and partial ring PET scanners (#871)

M. Béguin1, V. Commichau1, J. Flock1, C. Fuentes1, T. Lomax2, S. Makkar2, K. McNamara2, M. Oddo3, J. O. Prior4, C. Ritzer1, D. C. Weber2, C. Winterhalter2, G. Dissertori1

1 ETH Zürich, Institute for Particle Physics and Astrophysics, Zürich, Zürich, Switzerland
2 Institut Paul Scherrer, Center for Proton Therapy, Villigen, Aargau, Switzerland
3 Lausanne University Hospital, Direction Médicale et Service de Médecine Intensive, Lausanne, Vaud, Switzerland
4 Lausanne University Hospital, Nuclear Medicine and Molecular Imaging Department, Lausanne, Vaud, Switzerland


The PETITION (PET for InTensive Care units and Innovative protON therapy) project aims to design two new bedside Positron Emission Tomography (PET) scanners for Intensive Care Units (ICU) and proton beam therapy applications. The first device is intended to scan deeply sedated, critically-ill patients with sepsis infection directly at the ICU, as their condition does no allow for a transportation to regular PET/CT facilities. The second device is designed for online image acquisition for in vivo dosimetry and hypoxia tracking studies. While the ICU device is a conventional full ring PET scanner, the other system has a U-shaped design ensuring a suitable opening for protons irradiation. The attenuation of photons is the most important limiting factor of quantitative accuracy and quality of PET images. To perform attenuation correction, the knowledge of the patient-specific attenuation map (μ-map) is required. Several approaches, using either external information or emission data, already exist for the μ-map generation. The PETITION scanners do not offer the possibility to perform a transmission scan and therefore the attenuation map needs to be generated from emission data only. While the transmission-less method is already widely applied with full ring PET scanners,it is usually not used with partial ring devices. This document presents preliminary results of attenuation correction based on the emission of full and partial ring PET data only.

AcknowledgmentThe authors would like to thank the Swiss National Science Foundation for supporting the PETITION project under grant CRSII5_189969.
Keywords: Positron Emission Tomography (PET), Partial ring PET scanner, Attenuation correction, Transmission-less, Quantitative imaging

Locating Radio-frequency Coils for Inclusion in MR-based PET Photon Attenuation Correction in Simultaneous PET/MRI (#1091)

E. Anaya1, 2, P. Schleyer1, C. S. Levin3, 2

1 Siemens Medical Solutions USA, Inc., Siemens Healthineers, Molecular Imaging, Knoxville, Tennessee, United States of America
2 Stanford University, Electrical Engineering, Stanford, California, United States of America
3 Stanford University, Radiology, Stanford, California, United States of America


In simultaneous positron emission tomography and magnetic resonance (PET/MR) imaging, MR radio frequency (RF) coils are placed on the patient to receive the MR signal. These coils can produce an undesirable attenuation of the PET signal by as much as 17% in certain cases. Currently, photon attenuation correction (AC) of flexible RF MR coils is not typically performed in commercial PET/MR systems. To correct for this attenuation, the position of the coils must be determined. This work proposes a simple and effective solution to this problem by using three 2D optical cameras placed just outside the field-of-view (FOV) of the PET/MR to determine the position and orientation of markers attached to RF coils. An average marker location error of 7.7 mm was achieved when identifying 8 markers on a flexible RF coil placed on a phantom. Experimental quantification of reconstructed PET signal error due to inaccurate assessment of flexible RF coil location on a phantom is presented. Given the location accuracy of this optical method, the PET signal attenuation error is reduced to less than 3%. Our method can be extended to correct for other attenuating hardware objects in the FOV of PET/MR.

AcknowledgmentWe would like to thank Judson Jones and Siemens Healthineers in Knoxville, USA for help with the project. This work was partially supported by the National Science Foundation under Grant No. 1828993.
Keywords: PET/MR, attenuation correction, optical camera, optical tracking, radio-frequency (RF) surface coils

Deep learning-based attenuation correction strategies in the sinogram domain (#1191)

H. Arabi1, H. Zaidi1

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


This work focuses on the investigation of the different deep learning-based approaches for the task of PET AC in the sinogram space. In this regard, different deep learning models were developed for non-time of flight (nonTOF) and TOF PET attenuation correction using the emission data in the sinogram domain. These models include direct estimation of the emission sinograms corrected for attenuated photons (AC) from the uncorrected/raw sinograms (nonAC), ACFs estimation from nonAC sinograms, independent development of the deep learning models for each sinogram segment, and scatter correction of the nonAC sinograms prior to the development of the models. A segmentation-based 2-class AC map was also evaluated in this study to provide a bottom-line for performance assessment of the different models. This study included 70 clinical TOF PET/CT brain examinations. The resulting PET images from the different AC models were quantitively assessed through calculation of the region-wise standardized uptake value (SUV) bias. Overall, the AC models relying on time-of-flight information exhibited superior performance to the non-TOF models (as well as 2-class AC approach) with SUV biases of 6.5% (TOF model), 9.5% (non-TOF model), and 14.0% (2-class). Incorporation of the prior scatter correction in the model for ACFs estimation was critical (for both TOF and non-TOF models). On the other hand, estimation of the attenuation corrected sinograms directly from the non-AC sinograms showed no sensitivity to the prior scatter correction. Regarding TOF data, direct estimation of the AC sinograms does not need prior correction for scatter photons, however, its implementation is computationally expensive. On the other hand, estimation of the ACFs from TOF sinograms is less memory demanding, however, this model requires prior scatter correction to reach peak performance.

Keywords: PET, sinogram, deep learning, attenuation correction, quantitative imaging

An Integrated framework of projection and attenuation correction for quantitative SPECT/CT reconstruction (#1336)

L. Cheng1, 2, F. Liu3, L. Gao3, L. Sun4, Y. Hou3, Y. Liu1, 2

1 Tsinghua University, Department of Engineering Physics, Beijing, China
2 Tsinghua University, Key Laboratory of Particle & Radiation Imaging, Ministry of Education, Beijing, China
3 Beijing NOVEL MEDICAL Equipment Ltd., Beijing, China
4 CNNC High Energy Equipment(Tianjin) Co. Ltd, Tianjin, China


Quantitative SPECT imaging has become more common due to increasing demands in clinical [1]. With the development of SPECT/CT, quantification techniques for SPECT have also developed rapidly in recent years. However, the performance of the existing methods, such as reconstruction speed and quantification accuracy, needs further optimization and validation. In this work, we propose a new framework which has the potential to improve the reconstruction speed and accuracy simultaneously. Experimental studies were conducted to evaluate the proposed method. For a typical 64×64×64 projection data, one iteration reconstruction costed only 0.13 s. Quantification accuracy evaluation based on a uniform cylinder phantom demonstrated a relative error of 6.3%. At last, a Jaszczak phantom was employed for quality control purpose. The reconstructed results showed similar performance to those from clinical work station.  

Keywords: quantitative SPECT, attenuation correction, multi-ray tracing

An image processing method for high-temporal-resolution dynamic PET imaging (#113)

Z. Chen1, Y. Duan2, Y. Wang3, C. Li3, D. Liang1, Y. Yang1, Z. Cheng2, Z. Hu1

1 Shenzhen Institute of Advanced Technology Chinese Acadamy of Science, Institute of Biomedical and Health Engineering, Shenzhen, China
2 The First Affiliated Hospital of Shandong First Medical University, Department of PET/CT, Jinan, China
3 Shanghai United Imaging Healthcare, Central Research Institute, Shanghai, China


Dynamic PET imaging (dPET), yielding multi-frame images, provides more comprehensive information than conventional static PET in terms of the temporal evaluation of the tracer and the metabolic parameters of patients from subsequent parametric analysis. High temporal resolution (HTR) is desired for the early frames because of the rapid variation of the tracer concentration; however, this is limited by the imaging quality with short frame durations. Many advanced algorithms for better imaging quality with higher temporal resolution have been proposed, but these methods are inconvenient for industrial applications since a procedure of image reconstruction from law data or a complicated network model that requires training is involved. In this work, we propose a dPET image optimization algorithm based on the application of third order Hermite interpolation to improve the imaging quality of early time frames with higher temporal resolution. Simulative and clinical 2-meter dPET image data were used to validate the proposed method. The experimental results fully demonstrated the superior performance in high-temporal-resolution (HTR) dPET imaging of the proposed image optimization algorithm.

AcknowledgmentThis work was supported by the National Natural Science Foundation of China (32022042, 81871441). 
Keywords: Dynamic PET, Image processing, Temporal resolution, Third order Hermite interpolation

Blood Input Function Estimation in Positron Emission Tomography with Deep Learning (#248)

L. Szirmay-Kalos1, D. Varnyú1

1 Budapest University of Technology and Economics, Budapest, Hungary


Dynamic positron emission tomography allows in vivo study of metabolic processes by monitoring the tissue uptake of a radioactive tracer. Its aim is to reconstruct the time-activity curves of the voxels of the measured volume, which are sought in the algebraic form of a predetermined kinetic model. This kinetic model typically depends on the blood input function, which describes the amount of radiotracer in the blood that can be absorbed by tissues. The blood input function can be estimated in parallel with the kinetic parameters of the voxels during the iterative reconstruction process. However, because kinetic parameter calculation requires the blood input even at the start of reconstruction, an initial approximation needs to be given. In this paper, a deep CNN-LSTM-DNN network is proposed to estimate the blood input function from the sinogram data.

AcknowledgmentOTKA K-124124
Keywords: PET, Dynamic tomography, Blood input function, deep learning

Head Motion Correction Using an EPI-Based Method in a [11C]ABP688 PET/MR Study (#296)

C. R. Brambilla1, 2, O. Zeusseu1, 4, J. Scheins1, E. R. Kops1, L. Tellmann1, E. Farrher1, N. J. Shah1, 3, I. Neuner1, 2, C. Lerche1

1 Forschungszentrum Jülich GmbH, Institute of Neuroscience and Medicine, INM-4 / Imaging Physics, Jülich, North Rhine-Westphalia, Germany
2 RWTH Aachen University, Department of Psychiatry, Psychotherapy and Psychosomatics, Aachen, North Rhine-Westphalia, Germany
3 RWTH Aachen University, Department of Neurology, Aachen, North Rhine-Westphalia, Germany
4 HTW Saarlandes University, Department of Medical Physics, Saarbrücken, Saarland, Germany


Dynamic multimodal acquisitions with [11C]ABP688 have been used to study schizophrenia[4]. However, it is challenging for the subject to stay still during the data acquisition. In addition, head motion is a source of image degradation that affects quantification. Here, an implementation of the multiple acquisition frame (MAF)[5] to reduce the effect of this problem using echo-planar imaging (EPI) sequences already obtained from the original protocol for motion tracking is described, and preliminary data are shown. Head motion correction was performed using EPI sequences scattered along the originally acquired study protocol. Our approach includes a small number of the same EPI sequence for motion tracking around the existing functional and spectroscopy sequences and for attenuation map position tracking. A MATLAB[1] script was developed, which included SPM[2] functions, to extract motion information from the selected EPI scans. Parameters were converted into the PMOD[3] coordinate system and exported to an already existing transformation matrix file in the PET space. The matrix was then applied to a point cloud image that transfers the motion parameters to the attenuation map during a reconstruction step. Finally, following the MAF reconstruction step, a final realignment was performed, resulting in a dynamic PET image corrected for head motion. As a result, the time-activity curves present less variability, an increase in the measured mean activity concentration in the brain cortex from 5.08±0.03 non-corrected to 5.43±0.01 corrected with the EPI-based method can be seen, and there is a reduced coefficient of variance. The additional EPI scans for position tracking result in an almost negligible increase in MR acquisition time and can be integrated around the existing MR sequences.

[1] //
[4] Eur Psy, v50, Apr2018.
[5] IEEE Trans Med Im, v16, Apr1997.


EU-FP7 TRIMAGE (602621) and DAAD (57299294).

Keywords: Dynamic PET, EPI images, head motion correction, multimodal study, PET quantification

Knowledge distillation: A strategy to enhance the performance of deep learning-based seminal segmentation (#368)

R. Karimzadeh1, H. Arabi2, E. Fatemizadeh1, H. Zaidi2

1 sharif university of technology, Electrical engineering, Tehran, Iran (Islamic Republic of)
2 Geneva University Hospital, Department of Medical Imaging, Geneva, Genève, Switzerland


Accurate segmentation of target tissues/structures as well as surrounding healthy organs/tissues (organs at risk (OARs)) plays a critical role in radiation therapy treatment planning. Accurate segmentation of OARs prevents/minimizes unwanted toxicity to healthy tissues. Manual segmentation is time-consuming, tedious, prone to human errors and subject to intra- and inter-observer variability. In this regard, deep learning algorithms have shown extraordinary performance in automated organ segmentation from medical images, though the powerful/highly effective models might be computationally intensive. Transferring knowledge from complex/cumbersome models to simple/versatile models, known as knowledge distillation, has been proposed to address this issue (enhance the performance of the existing deep learning models). In this work, the impact of the knowledge distillation on OARs segmentation from CT images is investigated for commonly used Unet and Resnet deep learning models. To this end, a highly complex Unet model (as teacher) and two conventional deep learning models (Unet and Resnet as students) were developed to delineate OARs on thoracic CT images from the SegTHOR public dataset. The models were trained once independently and once through knowledge distillation from the teacher model to the student models. The teacher model yielded segmentation accuracy in terms of Dice coefficient of 0.95 and 0.86 for the heart and aorta compared to the student models (Unet and Resnet) which achieved an accuracy of 0.79, 0.64 and 0.13, 0.22, respectively.  After knowledge distillation from the teacher to the students, the accuracy of the Unet and Resnet improved to 0.91, 0.79 and 0.62, 0.63 for the heart and aorta, respectively. This study demonstrated the beneficial impact of knowledge distillation to enhance the overall performance of conventional models without increasing the computational their complexity.

AcknowledgmentThis work was supported by the Swiss National Science Foundation under grant SNRF 320030_176052.
Keywords: knowledge distillation, organs at risk segmentation, deep learning, medical image segmentation

Posterior Estimation for Dynamic PET imaging using Conditional Variational Inference (#417)

X. Liu1, 2, T. Marin1, 2, A. Tiss1, 2, J. Woo1, 2, G. El Fakhri1, 2, J. Ouyang1, 2

1 Massachusetts General Hospital, Gordon Center for Medical Imaging, Boston, Massachusetts, United States of America
2 Harvard Medical School, Department of Radiology, Boston, Massachusetts, United States of America


This work aims efficiently estimating the posterior distribution of kinetic parameters for dynamic positron emission tomography (PET) imaging given a measurement of time of activity curve. Considering the inherent information loss from parametric imaging to measurement space with the forward kinetic model, the inverse mapping is ambiguous. The conventional (but expensive) solution can be the Markov Chain Monte Carlo (MCMC) sampling, which is known to produce unbiased asymptotical estimation. We propose a deep-learning-based framework for efficient posterior estimation. Specifically, we counteract the information loss in the forward process by introducing latent variables. Then, we use a conditional variational autoencoder (CVAE) and optimize its evidence lower bound. The well-trained decoder is able to infer the posterior with a given measurement and the sampled latent variables following a simple multivariate Gaussian distribution. We validate our CVAE-based method using unbiased MCMC as the reference for low-dimensional data (a single brain region) with the simplified reference tissue model.

Keywords: Dynamic PET imaging, Posterior Estimation, Variational Inference, Kinetic Model

Virtual High-Count PET Image Generation using a Deep-Learning Method (#532)

J. Liu1, S. Ren1, N. Mirian1, Y. - J. Tsai1, D. Pucar1, M. - K. Chen1, C. Liu1

1 Yale university, Department of Radiology and Biomedical Imaging, New Haven, Connecticut, United States of America


The work applied deep learning to reduce the noise of standard-dose PET images to obtain virtual-high-statistics PET images. The training datasets, including down-sampled PET images as the network input and full dose images as the network output, were derived from 27 PET datasets, each acquired using 90-min scan time. The trained network was evaluated on 195 clinical data. The results showed that the virtual-high-count images have reduced noise level, and there is no significant difference on mean/max SUV between the standard-dose and virtual-high-dose PET images.

Keywords: virtual-high-count, denoising, deep learning

Investigation of noise reduction in low-dose SPECT myocardial perfusion images with a generative adversarial network (#639)

N. Aghakhan Olia1, A. Kamali Asl1, S. Hariri Tabrizi1, P. Geramifar2, P. Sheikhzadeh3, H. Arabi4, H. Zaidi4

1 Shahid Beheshti University, Department of Medical Radiation Engineering, Tehran, Iran (Islamic Republic of)
2 Tehran University of Medical Sciences, Research Center for Nuclear Medicine/Shariati Hospital, Tehran, Iran (Islamic Republic of)
3 Tehran University of Medical Sciences, Department of Nuclear Medicine/ImamKhomeini Hospital Complex, Tehran, Iran (Islamic Republic of)
4 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Switzerland


The purpose of this work was to investigate the feasibility of dose reduction while preserving crucial structures and clinical value in SPECT-MPI images. In this study, we acquired 330 clinical SPECT-MPI images from a dedicated cardiac SPECT in list-mode format. All patients underwent a two-day rest/stress protocol and the obtained non-gated images were retrospectively used to convert low-dose images to normal-dose ones in the projection space. A deep generative adversarial network was employed to predict normal-dose images from 50%, 25%, and 12.5% of the normal-dose level in the projection space. Peak signal-to-noise ratio (PSNR), root mean squared error (RMSE), and structural similarity index (SSIM) were used to evaluate the performance of the proposed deep learning-based denoising model at different reduced dose levels. Moreover, a Pearson correlation coefficient analysis was performed on derived parameters from QPS package. The quantitative analysis demonstrated that the highest PSNR (42.49 ± 2.37) and SSIM (0.99 ± 0.01), and the lowest RMSE (1.99 ± 0.63) were obtained at the half-dose level measured in the reconstructed images. The quantitative analysis demonstrated that the deep learning model could effectively recover the underlying information in 1/2-dose and 1/4-dose SPECT images. However, due to the extremely high noise level in 1/8-dose, the proposed network failed to suppress the noise and recover underlying structures successfully.

Keywords: SPECT, Myocardial Perfusion Imaging, Denoising, Low-dose, Deep Learning

Evaluation of Data Driven Respiratory Signal Extraction Methods from Cone-Beam CT using MR-based Digital Phantoms (#696)

A. T. Mohd Amin1, S. S. Mokri1, R. Ahmad2, A. A. Abd Rahni1

1 Universiti Kebangsaan Malaysia, Faculty of Engineering and Built Environment, Bangi, Malaysia
2 Universiti Kebangsaan Malaysia, Faculty of Health Science, Kuala Lumpur, Malaysia


A comparative study to assess the performance of data-driven methods to extract respiratory signals from Cone-Beam CT (CBCT) projection data was earlier implemented using a recently developed 4D MRI-based CBCT digital phantom dataset. However, a more extensive study is deemed necessary to assess the accuracy and robustness of each method rather than just rely on a single dataset. Thus, the purpose of this work is to further investigate the methods using the developed 4D MRI-based CBCT digital phantom datasets from five volunteers with actual different anatomical and breathing features. The effect of a wider range of justified region-of-interest (ROI) projection data range selection for each algorithm is also considered to further assess these methods. The results indicate that the Local Principal Component Analysis (LPCA) manage to reproduce its highest accuracy with an overall mean correlation coefficient value at 0.815 compared to Amsterdam Shroud (AS), Intensity Analysis (IA) and Fourier Transform (FT) methods. The LPCA method also demonstrated robustness with a highest correlation across different range of ROI as long as the diaphragm is present, though dissimilar to the high sensitivity of the FT method that is susceptible to only the diaphragm and not the presence of other organs.

Keywords: Cone beam CT, data driven, digital phantom, respiratory signal

Does prior knowledge in the form of multiple low-dose PET images (at different dose levels) improve standard-dose PET prediction? (#775)

B. Sanaei1, R. Faghihi1, H. Arabi2, H. Zaidi2

1 Shiraz University, Nuclear Engineering Department, Shiraz, Iran (Islamic Republic of)
2 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Genève, Switzerland


To acquire a high-quality PET image, a standard dose of radioactive tracer is injected into the patient which may pose a high risk of radiation exposure damage. On the other hand, reducing the injected dose increases the statistical noise in the PET images. To improve the image quality of low-dose PET (L-PET) images, deep learning methods have been introduced to denoise the L-PET images, wherein the relationship between the L-PET images and the standard-dose PET (S-PET) images is learned by the model to predict the S-PET images from their low-dose counterparts. The existing deep learning-based approaches solely focus on a single level of L-PET imaging to predict the S-PET images. In this work, we investigate the benefits of exploiting multiple PET images at lower dose levels (in addition to the target low-dose level) as prior knowledge to predict the S-PET images. To this end, a high-resolution residual deep learning network was employed to develop two S-PET prediction models. First, the network was trained using a single input channel for 8% L-PET images. In the second model, multiple L-PET images (6% and 4%, in addition to 8% L-PET) were considered as inputs to the network. The performance of the two models was evaluated using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), root mean square error (RMSE) of standard uptake value (SUV) within the entire head region. Moreover, mean SUV bias (SUVmean) was calculated for the malignant lesions. The quantitative analysis of 20 patients in the external validation dataset demonstrated the superior performance of the multi-input model. The RMSE within the entire head region reduced from 0.12±0.04 in 8% L-PET, to 0.09±0.03 and 0.06±0.02 for the single- and multi-input models, respectively. Moreover, the SUVmean bias reduced from -4.18±1.14% in the single-input model to -1.44±0.56% in the multi-input model. This study demonstrated the benefits of using multiple L-PET images to estimate the S-PET images.

Keywords: PET, Quantitative imaging, Low dose, Deep learning

Neural Networks for event classification of Compton Camera data from multi-energy radioactive sources (#823)

P. Martín-Luna1, L. Barrientos1, M. Borja-Lloret1, E. Muñoz1, A. Ros1, J. Roser1, R. Viegas1, V. Sanz1, G. Llosá1

1 Instituto de Física Corpuscular (CSIC-U. Valencia), Paterna, Spain


Background rejection in Compton cameras can result in a substantial improvement of the reconstructed images. The IRIS group of IFIC – Valencia is employing neural networks for this purpose in the development of Compton cameras for different applications. Following the initial studies in hadron therapy monitoring, the technique is being applied to low energy radioactive sources for medical imaging. The neural network is trained with simulated data and then employed for event selection prior to image reconstruction. The hyperparameters
have been optimised in order to obtain the best classification for the simulated data. Results are compared to energy cuts selection, both in the precision of event classification and in the subsequent reconstructed images showing an improvement both with simulated and real data.

Keywords: Event classification, Compton camera, background reduction, Neural Network

Delay calibration for ultrasound computed tomography system using a neural network (#888)

N. Shen1, H. Wang1, X. Lei2, S. Yu1, X. Xia1

1 Donghua University, School of Computer Science and Technology, Shanghai, China
2 Zhejiang Equilibrium Nine Medical Technology Company, Hangzhou, China


System errors of ultrasound computed tomography system will affect the quality of image reconstruction. Among these errors, the transducer delay has the largest magnitude, bringing the greatest harm to image reconstruction. In this paper, we propose a neural network method for transducer delay calibration. The method transforms the calibration problem into solving a large-scale system of linear equations, and uses a neural network to find the optimal solution, where the solution consists of the approximate value of each transducer’s delay. We test the method with simulated system data where we add transducer delays in the range of 0.7~1.3 μs. We compare seven optimizers for training the neural network in order to find the best solution with the smallest errors. Experimental results show that the errors of calibrated results can be reduced under 0.3×10-3 μs, significantly outperforming existing methods.

AcknowledgmentThis work was supported by “the Fundamental Research Funds for the Central Universities” from Donghua University under Grant No. 2232020D-36, and “the Young Teacher Research Startup Fund” from Donghua University under Grant No.112-07-0053079.
Keywords: 3D USCT system, system calibration, neural network

CNN-based time delay estimation in dynamic total-body PET kinetic modeling (#910)

E. Li1, E. Berg1, Y. Wang2, B. A. Spencer1, R. D. Badawi2, A. F. Tarantal3, 4, G. Wang2, S. R. Cherry1

1 University of California Davis, Department of Biomedical Engineering, Davis, California, United States of America
2 University of California Davis School of Medicine, Department of Radiology, Sacramento, California, United States of America
3 University of California Davis School of Medicine, Departments of Pediatrics and Cell Biology and Human Anatomy, Davis, California, United States of America
4 University of California Davis, California National Primate Research Center, Davis, California, United States of America


Time delay correction of the input function for quantitative kinetic modeling is necessary to reduce errors in parameter estimates, particularly when incorporating arterial blood sampling systems, or selecting an image-derived input function (IDIF) far from the tissue of interest, where large sources of external and internal delay respectively must be corrected. Delay can be jointly estimated during the fitting process; however, for long axial field-of-view scanners, delay correction to tissue voxels is computationally prohibitive, particularly for parametric imaging of total-body dynamic PET data. We propose an efficient solution to estimate the delay time between an IDIF derived from the left ventricle to individual tissue voxels using convolutional neural networks (CNN). We demonstrate the improvement in parameter bias from the CNN-based delay estimates. Further we show these results across two tracers and two different total-body PET systems.


This work was supported by NIH R01 CA206187, NIH R35 CA197608 and OD11107.

Keywords: blood flow, convolutional neural network (CNN), time delay, kinetic modeling, total-body PET

Deep learning-based Dosimetry in Radionuclide Therapy: Is It Worth the Effort? (#945)

A. Akhavanallaf1, Y. Salimi1, I. Shiri1, H. Arabi1, X. Hou2, J. M. Beauregard3, A. Rahmim2, H. Zaidi1

1 geneva university, Department of Radiology & Medical Informatics Geneva University Hospital, Department of Diagnostics, Geneva, Switzerland
2 University of British Columbia, Department of Radiology, Vancouver, Canada, Vancouver, Canada
3 Université Laval, Cancer Research Centre and Department of Radiology and Nuclear Medicine, Quebec, Canada


We propose a novel unified framework to perform whole-body voxel-level dosimetry taking into account patient-specific tissue heterogeneity and activity distribution using Monte Carlo (MC) simulations and deep learning algorithms. We extended the core idea of the voxel-scale MIRD dosimetry formalism previously validated for positron emitters used in diagnostic imaging (18F) to radionuclides with complex decay schemes used in therapy (177Lu). In this context, we trained a model to predict the deposited energy distribution obtained from MC simulations (specific S-values), while two-paired input channels consist of density map and dose distribution kernel in soft-tissue (single S-value) are fed into the network. Transfer learning was applied using our previous 18F model fine-tuned on 177Lu dataset. Accordingly, whole-body dose maps were constructed through convolving specific S-values into time-integrated activity distribution obtained from SPECT images. The Deep Neural Network (DNN) predicted dose map was compared with the reference (Monte Carlo-based) and two MIRD-based methods, including single-voxel S-value (SSV) and multiple voxel S-value (MSV) approaches. The results demonstrated that DNN, MSV and SSV show a comparable performance against the MC approach in soft tissue. However, in small size heterogeneous boundaries (lumbar region), DNN outperformed other approaches achieving lower bias (4%) compared to MSV (26%) and SSV (30%) tech.


This work was supported by the Euratom research and training program 2019-2020 Sinfonia project under grant agreement No 945196 and Iran’s Ministry of Science.

Keywords: Internal dosimetry, deep learning, radionuclide therapy, Monte Carlo

Lung Cancer Recurrence Prediction Using Radiomics Features of PET Tumor Sub-Volumes and Multi-Machine Learning Algorithms (#965)

S. A. Hosseini2, G. Hajianfar3, I. Shiri1, H. Zaidi1

1 Geneva University Hospital, Division of Nuclear Medicine and Molecular Imaging, Geneva, Genève, Switzerland
2 Tehran University of medical sciences, Department of Medical physics and biomedical engineering, Tehran, Iran (Islamic Republic of)
3 Iran University of Medical Science, Rajaie Cardiovascular Medical and Research Center, Tehran, Iran (Islamic Republic of)


We aimed to predict recurrence in lung cancer patients using PET radiomics features and machine learning algorithms. In this work, 136 non-small cell lung cancer (NSCLC) patients were enrolled. To study the impact of tumor sub-volume on recurrence prediction's accuracy, five sub-regions or contours were delineated manually, were extended with different distances (1, 2, 3, 4, and 5 mm). Three different feature selections and ten classifiers with 100 bootstraps were utilized. Our results illustrated that contourPlus1mm with the Minimum Redundancy Maximum Relevance (mrmr) feature selection and Random forest (RF) classifier, contourPlus1mm with the MRMR feature selection and Linear Discriminant Analysis (LDA) classifier, and contourPlus4mm with the Recursive Feature Elimination (RFE) feature selection and Logistic regression (LR) classifier, had the highest performance (AUC= 0.65). The results of this study illustrated that an extended sub-volume of a manual contour boosts the performance of recurrence prediction in patients with lung cancer. This study demonstrated that the use of contour extending method can be effective in increasing the predictive accuracy of different machine learning classifier methods.

AcknowledgmentThis work was supported by the Swiss National Science Foundation under grant SNRF 320030_176052; the Swiss Cancer Research Foundation under Grant KFS-3855-02-2016
Keywords: Radiomics, Machine learning, Lung cancer, Prognostic, Recurrence

An efficient end-to-end Convolutional Neural Network for classification of diabetic retinopathy using RestNet (#1046)

F. A. A. Slimani1, M. Bentourkia1

1 University of Sherbrooke, Nuclear Medicine and Radiobiology, Sherbrooke, Québec, Canada


Diabetes is a chronic disease affecting millions of people worldwide, and more than 25% of them have diabetic retinopathy (DR). DR is the leading cause of blindness in adults. DR is usually diagnosed with a biomicroscopic examination of the fundus after pupillary dilation, supplemented by fundus photographs. The challenge lies in the interpretation of these images by a physician specialist. In this work we propose a computer-assisted pipeline for the diagnosis of DR using the convolutional neural network (CNN). This pipeline has three main stages: (i) image preprocessing, (ii) feature extraction and (iii) classification. First, we used the Discrete Wavelet Method (DWT) for edge segmentation, and the Contrast Limited Adaptive Histogram Equalization (CLAHE) method to improve image contrast. In the second and third step we used the ResNet50 convolutional neural network for the detection and classification of five levels of disease severity (0 - No DR, 1 - Mild, 2 - Moderate, 3 - Severe, 4 - Proliferative DR) on 35 122 images from the publicly available Kaggle database. To analyze the performance of the learning model, different metrics were used to detect different diseases, such as sensitivity, specificity, and the confusion matrix which includes the number of true positive, false positive, true negative and false negative. In conclusion, we achieved 85% accuracy in the validation phase and 81% in the testing phase, demonstrating the reliability of CNNs to identify and automate the diagnosis of diabetic retinopathy using digital fundus images.

Keywords: Convolutional Neural Networks (CNN), Deep learning, Image Classification, Discrete wavelet transform, Diabetic retinopathy

Deep learning-based low-dose cardiac gated SPECT: in projection or image domain? (#1089)

N. Aghakhan Olia1, A. Kamali Asl1, S. Hariri Tabrizi1, P. Geramifar2, P. Sheikhzadeh3, H. Arabi4, H. Zaidi4

1 Shahid Beheshti University, Department of Medical Radiation Engineering, Tehran, Iran (Islamic Republic of)
2 Tehran University of Medical Sciences, Research Center for Nuclear Medicine/Shariati Hospital, Tehran, Iran (Islamic Republic of)
3 Tehran University of Medical Sciences, Department of Nuclear Medicine/Imam Khomeini Hospital Complex, Tehran, Iran (Islamic Republic of)
4 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Switzerland


Radiation exposure reduction in SPECT-MPI is critically important. However, lowering injected dose would lead to degradation of the diagnostic accuracy of this modality. In this study, we acquired a total of 335 clinical gated SPECT-MPI images from a dedicated cardiac SPECT in list-mode format. All patients underwent a two-day rest/stress protocol and the obtained gated images were retrospectively used to convert low-dose data to normal-dose data in both projection and image spaces. A deep generative adversarial network was employed to predict normal-dose images from 50% low-dose data. As a measurement relevant to the overall efficiency of the proposed network, the peak signal-to-noise ratio (PSNR), root mean squared error (RMSE), and structural similarity index measure (SSIM) quantitative metrics were measured. Moreover, a Pearson correlation coefficient analysis was performed on the half-dose and predicted normal-dose images with respect to the normal-dose images. The results demonstrated that the highest PSNR (46.30 ± 2.23) and SSIM (0.98 ± 0.01), and the lowest RMSE (1.32 ± 0.54) were obtained from the image space implementation. Pearson analysis showed that the predicted normal-dose data yielded ρ = 0.960 ± 0.011 and ρ = 0.947 ± 0.027 in the image and projection spaces, respectively. Overall, considering the quantitative metrics, the noise was effectively suppressed in the predicted normal-dose images in both spaces, however normal-dose data estimation in the image space resulted in superior quantitative accuracy and image quality.

Keywords: SPECT, Myocardial Perfusion Imaging, Denoising, Low-dose, Deep Learning

Towards 2D Dosimetry using Monolithic Active Pixel Sensors and a Copper Grating (#1138)

J. Velthuis1, 3, C. De Sio1, J. Pritchard1, Y. Li1, R. Hugtenburg2, 1, L. Beck1

1 University of Bristol, School of Physics, Bristol, United Kingdom
2 University of Swansea, Medical School, Swansea, United Kingdom
3 University of South China, School of Nuclear Science and Technology, Hengyang, China


Higher energy and intensity X-ray radiotherapy treatments are coming into wider use, having the benefit of requiring fewer treatment fractions and fewer hospital visits per patient. However, small percentage errors in MLC positioning and dose become bigger problems with higher doses per fraction. Hence, real-time treatment verification becomes essential. Where devices downstream from the patient suffer from scattering in the patient, upstream devices can disturb the therapeutic beam. Here, a method is proposed for performing dosimetry using Monolithic Active Pixel Sensors, which can be made thin enough to disturb the beam by <1%. In order to calculate the dose to the tumour, a verification device needs to make a measurement of the photon field. Some photons will Compton scatter an electron in the silicon and generate a signal.  However, this signal is obscured by energy deposits from contamination electrons, originating from Compton scattering in the accelerator head and air. Often extensive build-up material is added to verification devices to reduce the electron contamination and enhance the photon signal. However, this leads to degradation of the beam intensity to the patient. Instead we propose using thin strips of 50 μm thick copper in a grating pattern and measuring the difference in the signal with and without it. The contamination electrons are relatively undisturbed by the presence of the thin copper strips and the photon signal generated via Compton scattering is enhanced. Hence the difference in the two signals mostly consists of energy deposits originating from the therapeutic photons. From this the dose to the patient can be derived. Using this technique, we show that the electron contamination signal can be reduced from 38% of the total signal to 2.6%. This allows to extract the photon signal only from the data and thus the dose to patient with a very thin upstream detector.


This work was supported by the STFC UK.

Keywords: Radiotherapy, Dosimetry, monolithic active pixel sensors

Segmentation of the hippocampus head and body: comparison of single annotator and multi-annotator (#1185)

H. Arabi1, H. Zaidi1

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


This work set out to assess the performance of the multi-annotator segmentation framework for hippocampus head and body segmentation. The multi-annotator segmentation framework employs a number of independent deep learning-based models with different loss functions, algorithms, and structures in order to perform organ segmentation independently on the input MR image. The outcome of each deep learning model is produced in the form probability map to be later fused by decision fusion algorithms or a deep learning-based combiner to generate a single binary map of the target organ/structure. The multi-annotator segmentation approach is intended to take advantage of the complementary information generated/provided by the different independent annotators (independent deep learning models). This framework would benefit from the synergistic effect of combining the different annotators’ decisions to reach a superior performance compared to each of the annotators alone. The performance of the proposed multi-annotator approach was investigated for the delineation of the hippocampus head and body from MR images. A number of different deep learning algorithms were implemented (including different architectures and loss functions). Overall, the proposed multi-annotator segmentation approach, which contained six independent deep learning models including the ResNet model, outperformed other segmentation approaches with Dice indices of 91.0±1.3 for the body and 91.1±1.3  for the head. While the best atlas-based approach led to Dice indices of 88.4±1.5 for the body and 88.5±1.5 for the head, and multi-view segmentation method resulted in Dice indices of 88.9±1.5 (body) and 89.0±1.4 (head). This work demonstrated that the proposed multi-annotator approach for seminal segmentation is able to show a performance superior to each of the annotators alone (independent deep learning models which are included in the multi-annotator approach).

Keywords: Deep learning, multi-annotator, segmentation, MRI, hippocampus

Deep learning-based estimation of functional connectivity from time series multi-Channel EEG for schizophrenia patients (#1207)

A. Khodabakhsh1, H. Arabi2, H. Zaidi2

1 Sharif University of Technology, Biomedical Signal and Image Processing Laboratory (BiSIPL), School of Electrical Engineering, Tehran, Iran (Islamic Republic of)
2 Geneva University Hospital, Division of Nuclear Medicine & Molecular Imaging, Geneva, Switzerland


Human brain, as a complex network, is affected by many mental disorders and neurodegenerative diseases. Brain functional and structural alteration could be captured by imaging modalities such as CT and MR imaging. However, these modalities have low sensitivity to properly capture the brain connectivity map as a biomarker for early diagnosis of neurodegenerative diseases. In this light, complimentary examination tools (such as EEG) are employed to estimate the brain functional connectivity (FC) map. To decode the brain FC map from EEG signals, conventional approaches rely on hand-craft feature extraction that would lead to suboptimal performance/effectiveness. In this light, this work set out to implement a novel deep neural network to jointly extract the brain FC maps and identify (based on the obtained FC map) the type of neurodegenerative disease from the patient's EEG signals. Due to the absence of the ground truth brain FC maps, the proposed approach extracts the patient-specific brain FC maps in an unsupervised fashion. To evaluate the performance of the proposed deep learning model, a publicly available dataset of EEG signals from healthy control and schizophrenia patients was employed. The proposed model exhibited an accuracy of 91% to classify schizophrenia patients. Moreover, the estimated brain FC maps for both healthy control and schizophrenia patients exhibited highly discriminative patterns to differentiate abnormalities from the healthy controls. The proposed unsupervised model, which is applicable to EEG and functional MR data, exhibited promising performance.

Keywords: Deep Learning, Fucntional Connectivity, EEG, Schizophrenia

vPET-ABC: Voxel-wise approximate Bayesian inference for parametric imaging of neurotransmitter release (#1232)

C. Grazian1, 2, G. Emvalomenos3, 4, G. Angelis4, 5, Y. Fan1, 2, S. R. Meikle3, 4

1 University of New South Wales, School of Mathematics and Statistics, Sydney, Australia
2 ARC Centre of Excellence fFor Mathematical and Statistical Frontiers, Sydney, Australia
3 The University of Sydney, Sydney School of Health Sciences, Sydney, Australia
4 The University of Sydney, Brain and Mind Centre, Sydney, Australia
5 National Imaging Facility, Sydney, Australia


We recently developed a method, called PET-ABC, that provides complete Bayesian statistical analysis of ROI-based dynamic PET data. The aim of this work was to extend the method to voxel-based analyses (vPET-ABC), with an initial focus on parametric imaging of neurotransmitter (NT) release in PET activation studies. In this study, the kinetic model used in conjunction with vPET-ABC was the linear-parametric neurotransmitter PET (lp-ntPET) model. This model, which incorporates time-varying kinetic parameters, describes the effect of neurotransmitter changes on dynamic receptor-ligand data during PET activation studies. The vPET-ABC pipeline produces reliable estimates of voxel-based parameters and their associated posterior probability density distributions. It also computes model probability at the voxel level, where the two alternative models under consideration are lp-ntPET, which allows for an activation, and the more parsimonious Multilinear Reference Tissue Model (MRTM) which does not. Initials results, obtained from a simulated GATE 4D [11C]raclopride scan with realistic noise, demonstrated that vPET-ABC can provide insightful information about NT release at the voxel level, including the reliability of parameter estimates, which is important for reliably identifying subtle activations in noisy PET data. We believe the technique will be equally useful in the analysis of total body dynamic PET data, where the applicability of a single kinetic model to all tissues cannot be assumed.

Keywords: PET, voxel-wise analysis, Bayesian statistics, ABC

Self-Guided and MR-Guided Deep-Learned Post-Reconstruction PET Processing (#1403)

G. Corda-D'Incan1, J. A. Schnabel1, A. J. Reader1

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


Reconstructed images for PET often have a high level of noise and low spatial resolution, when shorter scan times and reduced injected doses are used. Regularisation methods such as post-reconstruction smoothing can help improving image quality. Recently, neural networks have proved to be highly effective for this task by learning an ensemble of kernels. In post-processing of PET images, a high resolution MR image can also be used for guidance to further improve the final image quality. In this work, we investigate the impact of the input choice and the number of training samples used on a neural network's performance for PET post-reconstruction processing. To do so, six combinations of low-count PET and MR independent reconstruction outputs are fed into a 3-layer residual convolutional neural network. The network was trained using as input the last iteration of a conventional PET reconstruction, all the iterates from the PET reconstruction, only the final PET and MR estimates, all the PET estimates and the final MR, the final PET and all the MR estimates then finally all the iteration outputs of independent PET and MR reconstruction. The networks have been trained using a different number of training samples as well. The results obtained suggest that using all the intermediate reconstructions help the network to perform better when the training set size is limited. Furthermore, the gain in performance observed when the dataset size increases is higher for methods using all the intermediate reconstruction outputs. Future work will focus on training networks with higher number of training samples to confirm the trend observed and assessing the proposed method on 3D real data.

Keywords: Post-reconstruction, Deep Learning, PET reconstruction

Training and Evaluation of the Lung Cancer Prediction Convolutional Neural Network (#719)

S. Leem1, W. Wu2, C. Kang1, S. Wang1, H. Cha3, K. Lee1, P. Kinahan2

1 Korea University, Biomedical Engineering, Seoul, Republic of Korea
2 University of Washington, Radiology, Seattle, Washington, United States of America
3 Korea Atomic Energy Research Institute, Daejeon, Republic of Korea


Early detection of lung cancer is crucial to increase the survival rate of patients. While lung nodules can be detected through screening or incidental CT imaging, those that are classified as indeterminant pose a diagnostic dilemma as only a small percentage are malignant, and lung biopsies have a risk of harm. Improved stratification and risk assessment of indeterminant pulmonary nodules will improve patient outcomes. Methods: We evaluated a variation of the Lung Cancer Prediction Convolution Neural Network (LCP-CNN) model designed to detect malignancy in early-stage lung nodules. Since LCP-CNN model parameters are proprietary, we trained the LCP-CNN using 856 (before augmentation) benign and malignant nodule CT images from the publicly available Lung Image Database Consortium (LIDC) image collection. We then evaluated the model performance using 117 benign and malignant nodule CT images. Results: After pretraining, the LCP-CNN achieved a diagnostic accuracy of 0.96 as measured by the area under the curve (AUC) of the receiver operating curve (ROC). The false positive rate was 5.1% and the false negative rate was 2.6%. Conclusions: The LCP-CNN achieved a promising level of diagnostic accuracy using publicly available lung image data sets. The performance level is sufficient to warrant further validation with external cohorts.

AcknowledgmentThis work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. NRF-2019M2D2A1A02059221).
Keywords: LCP-CNN, LIDC, TCIA, Hybrid Labeling Method

Generative Adversarial Network “Steerability” for Brain PET Image Generation (#978)

J. Penning1, R. John1, H. Chandler2, P. Fielding1, C. Marshall1, R. Smith1

1 Cardiff University, Wales Research and Diagnostic Positron Emission Tomography Centre (PETIC), Cardiff, United Kingdom
2 Cardiff University, Cardiff University Brain Research Imaging Centre, Cardiff, United Kingdom


Positron Emission Tomography (PET) imaging of the brain with 18F fluorodeoxyglucose (FDG) is the most accurate in vivo method for investigating regional human brain metabolism. With the ability to provide an indication of the rate of glucose utilization by neuronal tissue that reflects neuronal and glial activity. FDG PET is therefore an established procedure for assessing dementia disorders.  Artificial Intelligent (AI) solutions to knowledge representation and reasoning have promised themselves as a paradigm shift in the interpretation, analysis and processing of medical images.  Generative models with the capability to synthesize medical images offer the promise of enhanced data augmentation in the presence of unbalanced training data for the purposes of improved classification and predictive models.  Very little attention has been given to the quantitative evaluation of the generated images and their realism and diversity to be used for data augmentation.  In this work we create a Wasserstein Generative Adversarial Network (GAN) for 18F FDG Brain PET image synthesis in the case of an Alzheimer’s dataset and explore the ability to steer the synthesized images by introducing perturbations to the latent space model using a Beta distribution with varying parameters of α and β.  It was observed that adjusting the Beta parameters the network was able to produce features with greater structured similarity index measure towards early mild cognitive impairment, than when a more uniform distribution is utilized.  This suggests that the input distribution can manipulate the output and act as generative “weights”, which will be a subject of further investigation.

Keywords: Artificial Intelligence, generative adversarial network, alzheimer's, PET imaging, Brain Imaging

Challenges in optimization of a stationary tomographic Molecular Breast Imaging system (#1021)

K. Erlandsson1, A. Wirth2, K. Thielemans1, I. Baistow2, A. Cherlin2, B. F. Hutton1

1 University College London, Institute of Nuclear Medicine, London, United Kingdom
2 Kromek Ltd, County Durham, United Kingdom


A prototype Molecular Breast Imaging (MBI) system is currently under development, motivated by the need of a practical low-dose system for use in patients with dense breast tissue, where conventional mammography is limited. The system is based on dual opposing CZT detector arrays and multi-pinhole collimators which allow for multiplexing in the projection data. We have performed optimisation of various design parameters based on either contrast-to-noise ratio (CNR) in the reconstructed images or area-under-the-localisation-receiver-operating-characteristics curve (LROC-AUC) obtained using the scan statistic model. The optimisations were based on simulated data, and the parameters investigated were pinhole size and opening angle, pinhole separation and collimator-to-detector separation. The two optimisation approaches resulted in similar design parameters, allowing for reconstruction of tomographic images with high CNR and lesion detectability, which can lead to a reduced dose or scan time as compared to planar MBI.

AcknowledgmentThe Institute of Nuclear Medicine is supported by the NIHR University College London Hospitals Biomedical Research Centre.  Kromek are supported by an Innovate UK grant (104296).
Keywords: CNR, CZT, LROC, multi-pinhole collimator, multiplexing

Phantom-based image quality evaluation of the new digital PET/CT uMI Panorama (#1361)

Z. Deng1, Y. Sun1, Y. Liu1, L. He1, Y. Wu1

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


The latest digital PET/CT scanners using silicon photomultiplier tubes (SiPM) in the detector can provide both higher sensitivity and improved spatial and timing resolution.   uMI Panorama which was the digital PET/CT with United Imaging Healthcare (China) has spatial resolution of about 2.9 mm, sensitivity of 20.0 cps/kBq, peak NECR activity of 440 kcps and TOF resolution beyond 300 ps. Here, the NEMA/IEC NU2 image quality (IQ) phantom and Clinical Trials Network (CTN) torso phantom were adopted to study the improved image quality from the improved scanner physical performance. After the data was collected in list-mode, series images were acquired with various time subsets and reconstruction image matrix and also the coefficient of variation (COV) and contrast recovery coefficient (CRC) were obtained and compared. The results suggested uMI Panorama has excellent lesion detectability and image quality. All the spheres in the both phantoms can be observed even with 0.5-min and 1-min scanning duration per bed. The minimal scan time per bed was found to be less than 2 minutes which satisfied the COV<15%. And the CRC fluctuation with 7-mm and 10-mm sphere is about 40% more or less as a function of time subsets and image matrix, which is larger than the other spheres in the CTN study.

Keywords: uMI Panorama, Phantom evaluation, image quality

Design of Readout Electronics for Dose Monitoring Detectors in Hadrontherapy (#97)

F. Canclini1, 2, I. D'Adda1, 2, L. Buonanno1, 2, M. Carminati1, 2, C. E. Fiorini1, 2

1 Politecnico Of Milano, Dipartimento di Elettronica, Informazione e Bioingegneria (DEIB), Milano, Italy
2 Istituto Nazionale di Fisica Nucleare (INFN), Sezione di Milano, Milano, Italy


Proton and ion beam therapy are rapidly gaining importance for tumor treatments. The intrinsic capability to deposit energy in highly localized regions of the tissues implies the need to monitor the Bragg peak position to reduce safety margins in treatment planning. This can be achieved by measuring secondary radiation emitted during the interaction of hadrons within the target. Detection of prompt-gamma rays is a promising monitoring technique as it provides an almost instantaneous signal after the beam interaction with the tissue. Challenges in terms of detection efficiency, background reduction, compactness and ease of use of the developed prototype, impose the development of a specific detector for this kind of measurements.Detectors based on LYSO scintillator arrays readout by silicon photomultipliers (SiPMs), front-end ASICs specifically designed to deal with the large current signals expected at high energies, and a FPGA-based DAQ that will allow a real-time measurement, for each event, of the hit position, of the energy deposited in the pixel and of the time of occurrence, are under development.In this work, we focus specifically on the design of the readout electronics for such detection systems, as these are quite specific for this kind of applications, with respect to traditional nuclear medicine detectors, especially in terms of high dynamic range.

Keywords: ASIC, FPGA, Hadrontherapy, prompt-gamma, SiPM

Optical imaging of proton mini-beams (#199)

S. Yamamoto1, T. Yabe1, T. Akagi2

1 Nagoya University, Nagoya, Japan
2 Hyogo Ion Beam Medical Center, Tatsuno, Japan


Proton therapy using mini-beams is a promising method to reduce radiation damage to normal tissue. However, distribution measurements of mini-beams are difficult due to their small structures. Since optical imaging is a possible method to measure high-resolution 2-dimensional dose distribution, we conducted optical imaging of an acrylic block during the irradiation of mini-beams of protons. Mini-beams were made from a proton pencil beam irradiated to 1-mm slits made of tungsten plate. During irradiation of the mini-beams to the acrylic block, we measured the luminescence of the acrylic block using a charge-coupled device (CCD) camera. With the measurements, we could obtain slit beam images that have slit shapes in the shallow area while they were uniform in their Bragg peaks, which was similar to the case of simulated optical images by Monte Carlo simulations. We confirmed that high resolution optical imaging of mini-beams is possible and provides a promising method for efficient quality assessment (QA) of mini-beams as well as research on mini-beam therapy.

Keywords: mini-beam, optical imaging, proton, CCD camera, acrylic block

Imaging and range estimations of prompt X-rays using YAP(Ce) camera during particle-ion irradiation to non-uniform phantoms (#231)

M. Kitano1, S. Yamamoto1, T. Yabe1, T. Akagi2, T. Toshito3, M. Yamaguchi4, N. Kawachi4

1 Nagoya University Graduate School of Medicine, Department of Integrated Health Science, Nagoya, Japan
2 Hyogo Ion Beam Medical Center, Department of Radiation Physics, Tatsuno, Japan
3 Nagoya Proton Therapy Center, Nagoya City West Medical Center, Department of Proton Therapy Physics, Nagoya, Japan
4 National Institutes for Quantum and Radiological Science and Technology, Takasaki Advanced Radiation Research Institute, Quantum Beam Science Research Directorate, Takasaki, Japan


 Low-energy X-ray imaging of prompt secondary electron bremsstrahlung X-rays (prompt X-rays) emitted during particle-ion irradiation is a promising method for range estimation. However, measurements have so far been conducted mainly for uniform phantoms of water or an acrylic block. Prompt X-ray imaging for non–uniform phantoms has not yet been extensively measured or evaluated with realistic conditions. Consequently, we conducted imaging of prompt X-rays using a pinhole YAP(Ce) camera during irradiation of protons as well as carbon ions to non-uniform acrylic phantoms with small cavities and then evaluated the images and estimated the ranges from the measured prompt X-ray images. The non-uniform acrylic phantom used for imaging had a cylindrical cavity with a 20-mm or 10-mm diameter in the phantom. During irradiation of protons or carbon ions, imaging of one of the phantoms was conducted using the pinhole YAP(Ce) camera with an air cavity as well as filling the cavity with an acrylic rod. For the phantom with a 20-mm-diameter cavity, the prompt X-ray images measured for both protons and carbon ions showed the shape of the cavity in the images, and the ranges could be estimated from the images. For the phantom with a 10-mm-diameter hole, although the shape of the hole could not be clearly observed, the ranges could also be estimated from the images. Furthermore, Monte Carlo simulated prompt X-ray images with different spatial resolution of the X-ray camera showed similar images to the measured images. We confirmed that prompt X-ray imaging of non-uniform phantoms using the pinhole YAP(Ce) cameras was possible and that prompt X-ray imaging is a promising approach for estimating the ranges for both protons and carbon ions, even for non-uniform phantoms.

Keywords: non-uniform phantom, YAP(Ce) camera, prompt X-ray, carbon-ion, proton

Development of a simultaneous imaging system to measure the optical and gamma ray images of Ir-192 source for high-dose-rate brachytherapy (#258)

J. Nagata1, S. Yamamoto1, Y. Noguchi2, T. Nakaya2, K. Okudaira2, K. Kamada3, A. Yoshikawa3

1 Nagoya University Graduate School of Medicine, Department of Integrated Health Science, Nagoya, Japan
2 Nagoya University Hospital, Department of Radiological Technology, Nagoya, Japan
3 Tohoku University, New Industry Creation Hatchery Center, Sendai, Japan


In high-dose-rate (HDR) brachytherapy, verification of the Ir-192 source’s position during treatment is needed because such a source is extremely radioactive. One of the methods used to measure the source position is based on imaging the gamma rays from the source, but the absolute position in a patient cannot be confirmed. To confirm the absolute position, it is necessary to acquire an optical image in addition to the gamma ray image at the same time as well as the same position. To simultaneously image the gamma ray and optical images, we developed an imaging system composed of a low-sensitivity, high-resolution gamma camera integrated with a CMOS camera. The gamma camera has a 1-mm-thick cerium-doped yttrium aluminum perovskite (YA1O3: YAP(Ce)) scintillator plate optically coupled to a position-sensitive photomultiplier (PSPMT), and a 0.1-mm-diameter pinhole collimator was mounted in front of the camera to improve spatial resolution and reduce sensitivity. We employed the concept of a periscope by placing two mirrors tilted at 45 degrees facing each other in front of the gamma camera to image the same field of view (FOV) for the gamma camera and the CMOS camera. The spatial resolution of the imaging system without the mirrors at 100 mm from the Ir-192 source was 3.2 mm FWHM, and the sensitivity was 0.28 cps/MBq. There was almost no performance degradation observed when the mirrors were positioned in front of the gamma camera. The developed system could measure the Ir-192 source positions in optical and gamma ray images. We conclude that the developed imaging system has the potential to measure the absolute position of an Ir-192 source in real-time clinical measurements.

Keywords: CMOS camera, gamma ray, HDR brachytherapy, Ir-192 source, scintillation camera

Prompt gamma-ray measurements with radiomarkers for in vivo range verification and dose enhancement in protontherapy (#621)

S. Escribano-Rodriguez1, S. Paschalis1, M. Xiao1, S. Heil1, I. Syndikus1, D. Watts1, G. Vallejo-Fernandez1

1 University of York, York, United Kingdom


Proton therapy is an emerging modality for cancer treatment that induces a better dose conformation, compared to traditional photon radiotherapy, reducing the damage to healthy structures and tissues nearby. We studied the combination of two techniques to improve proton therapy treatments: prompt gamma ray imaging and the use of radiotracers. The combination of both techniques allows to enhance the dose, while verifying the range of the protons with the detection of the gamma rays emitted by the radiomarkers used. This is demonstrated through the comparison of experimental in-beam measurements conducted at KVI-CART, with a proton beam of 66.5 MeV, and a MonteCarlo simulation model developed. The results obtained show the potential of radiomarkers to measure the range of the proton beam wile enhancing the dose in the tumoral region.


We are grateful to the KVI-CART staff that supported our experiment by providing a high quality beam and valuable scientific input throughout the experiment. We are also grateful to the staff of the mechanical and electronics workshop at the Department of Physics, University of York, who helped us develop the target and construct the apparatus used in this work, respectively, and the Viking Cluster service from the University of York high performance team.

This work was part-funded by the Wellcome Trust [ref: 204829] through the Centre for Future Health (CFH) at the University of York, the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 654002, the Royal Society under grant number RG160320, the University of York STFC IAA and by University of York EPSRC studentship.

Keywords: Proton therapy, Prompt-gamma ray imaging, Radiomarkers, Range verification, Dose enhancement

Compton Imaging Study for Dose Monitoring in Carbon Therapy (#790)

C. Huang1, Q. Zhang1, J. Wen1, C. Pei1, Y. Li1, Y. Yin1, H. Peng1

1 Lanzhou, School of Nuclear Science and Technology, Gansu, China


To achieve precise radiotherapy for carbon ion therapy, it is necessary to accurately monitor the dose distribution in the patient's body. A double-layer LYSO Compton Camera system is designed in Geant4 simulation and in laboratory, which performs image reconstruction of γ-point source, and monitors the dose distribution in carbon therapy. We bombard the PMMA target with a 200 MeV/μ carbon beam and reconstructed 4.439 MeV prompt γ-ray image with Compton Camera system. The difference between the 3D dose distribution and the reconstructed prompt γ-ray distribution is about 9.3%. Finally, the LYSO Compton imaging system was used to perform Compton imaging experiments on a 22Na point source with a diameter of about 3 mm, and the FWHM of the reconstructed image was 4.05 mm, which verifies the experimental feasibility of the Compton imaging method.

AcknowledgmentThis work is supported by the National Natural Science Foundation of China 11875156, the Gansu guiding program of Science Technology Innovation, China of 2018ZX-07 and Gansu Province Youth Science and Technology Foundation, China 18JR3RA364.
Keywords: Carbon ion therapy, Compton Imaging, Dose monitoring

BENEdiCTE (Boron Enhanced NEutron CapTurE) Gamma-Ray Detection Module (#913)

A. Caracciolo1, L. Buonanno1, 3, I. D'Adda1, 3, D. Di Vita1, 3, A. Chacon2, M. Kielly2, M. Carminati1, 3, M. Safavi-Naeini2, C. Fiorini1, 3

1 Politecnico di Milano, DEIB, Milano, Italy
2 Australian Nuclear Science and Technology Organisation (ANSTO), Sydeny, Australia
3 Istituto Nazionale di Fisica Nucleare, Sezione di Milano, Milano, Italy


We present a gamma-ray detection module for Neutron Capture Enhanced Particle Therapy (NCEPT). The system has been optimised for boron-10 neutron capture agents that can be used for dose enhancement in proton and heavy ion therapy. The goal of the module is to distinguish the photopeak at 478keV from the prompt-gamma emission resulting from the ion-target nuclear interactions. The module consists of a compact 64-channel module, with a large array of SiPM coupled to a 2”×2” cylindrical LaBr3:Ce scintillator crystal (63ph/keV conversion efficiency, 16ns decay time). The electronic front- end ASIC features low-noise processing of photodetector signals, while the pixellated SiPMs detector and individual readout allows for position sensitivity in the crystal. We have characterised the energy resolution of the system experimentally, demonstrating a state-of-the-art energy resolution (3.27% at 662 keV), together with the capability of the FPGA-based DAQ integrated in the module to deploy an external synchronization signal to the ion beam bunches in order to generate anti-coincidence windows. This feature provides a mechanism to distinguish and reject scintillation events from prompt gammas, enhancing the signal-to-background ratio of the spectrometer.

Keywords: Gamma-ray spectroscopy, Boron Neutron Capture Therapy, Silicon Photomultipliers, Ion therapy

Quantitative helium-beam radiography of an anthropomorphic head and neck phantom exclusively based on silicon pixel detectors (#1068)

M. Metzner1, 2, F. Kehrein1, 2, C. M. Knobloch1, 2, G. Echner1, A. Runz1, B. Ackermann4, 2, S. Brons4, 2, O. Jäkel1, 4, M. Martišíková1, 2, T. Gehrke3, 1

1 German Cancer Research Center (DKFZ), Department of Medical Physics in Radiation Oncology, Heidelberg, Baden-Württemberg, Germany
2 National Center for Radiation Research in Oncology (NCRO), Heidelberg Institute for Radiation Oncology (HIRO), Heidelberg, Baden-Württemberg, Germany
3 University Hospital Heidelberg, Department of Radiation Oncology, Heidelberg, Baden-Württemberg, Germany
4 Heidelberg Ion-Beam Therapy Center (HIT), Heidelberg, Baden-Württemberg, Germany


Using quantitative ion-beam radiography to verify treatment plans is a powerful way to reduce uncertainties in ion-beam radiotherapy. In contrast to an X-ray computed tomography scan or an X-ray projection, an ion-beam radiograph shows the line integral of the relative stopping power (RSP) of the tissue directly, which renders a more accurate calculation of the position of the Bragg peak inside of the patient feasible.
We present a quantitative helium-beam radiograph of an anthropomorphic CIRS 731-HN phantom acquired at Heidelberg Ion-beam Therapy center using helium ions with several different energies up to 197.01 MeV/u. The detection system employing the Timepix technology developed at CERN consists of six fully pixelated silicon detectors which are used to measure position and direction of single helium ions in front and back of the imaged object, their time of arrival as well as their energy deposition.
Calibration curves were generated first by measuring the energy deposition of ions which traversed plastic phantoms of well known water equivalent thickness (WET). These curves were then applied to a 48 mm x 24 mm radiograph of the CIRS 731-HN anthropomorphic head phantom, translating the energy deposition values to WET, which is the integral of the RSP along the beam direction.
Comparing the quantitative helium-beam radiograph to a digitally reconstructed radiograph, it becomes apparent that the imaging modality presented in this article is a promising technique offering sufficient spatial resolution and the possibility to measure the integrated relative stopping power directly.

AcknowledgmentThis work is partially funded by the German research council (DFG) under the contract no. JA 1687/11-1
Keywords: helium-beam radiography, anthropomorphic phantom, ion-beam imaging, silicon pixel detectors, particle therapy

Characterisation of a double-sized Timepix3 mini-tracker in a mixed nuclear fragment field for real-time treatment monitoring in carbon-ion radiotherapy (#1302)

M. Subramanian1, 2, L. Kelleter1, L. Ghesquiere-Dierickx1, 3, T. Gehrke1, 4, R. Felix-Bautista1, 5, G. Echner1, J. Jakubek6, M. Jakubek6, L. Marek6, 7, P. Soukup6, D. Turecek6, M. Martisikova1

1 German Cancer Research Centre DKFZ, Department of Medical Physics in Radiation Oncology, Heidelberg, Baden-Württemberg, Germany
2 Otto-von-Guericke-Universitaet Magdeburg, Faculty of Electro Engineering and Information Technology, Magdeburg, Saxony-Anhalt, Germany
3 Heidelberg University, Medical Faculty, Heidelberg, Baden-Württemberg, Germany
4 Heidelberg University Hospital, Department of Radiation Oncology, Heidelberg, Germany
5 Heidelberg University, Department of Physics and Astronomy, Heidelberg, Germany
6 Advacam s.r.o, Prague, Czech Republic
7 Charles University, Institute of Particle and Nuclear Physics, Faculty of Mathematics and Physics, Prague, Czech Republic


Charged nuclear fragments produced during carbon-ion radiotherapy have been proposed to potentially enable so far unachievable real-time treatment monitoring. Observing changes in the fragment track distribution could allow changes in the patient anatomy to be detected during radiotherapy treatment. For this purpose, a novel mini-tracker made of two double-sized Timepix3 hybrid semiconductor pixel detectors has been characterised in a typical clinical mixed radiation field. The allocation of four chips on a single readout interface together with its nanosecond time resolution, as well as the dead-time and noise-free readout allow the reconstruction of individual fragment tracks. The results of the characterisation help to improve the design of a detection system that will be used in an upcoming clinical trial.

Keywords: beam range monitoring, carbon-ion radiotherapy, hybrid semiconductor pixel detector, nuclear fragments, Timepix3

Development of the positron-emitting 127Cs tracer for non-invasive imaging of radiocesium dynamics in living animals and plants (#143)

N. Suzui1, Y. - G. Yin1, Y. Miyoshi1, K. Kurita2, N. Kawachi1

1 National Institutes for Quantum and Radiological Science and Technology (QST), Takasaki Advanced Radiation Research Institute, Takasaki, Japan
2 Japan Atomic Energy Agency, Materials Sciences Research Center, Tokai, Japan


The accident at Tokyo Electric Power Company’s Fukushima Daiichi Nuclear Power Station in March 2011 caused various radioactive materials to fall over a wide area. In particular, 137Cs, which has a half-life of 30 years, had a serious impact on the affected area. Subsequently, interest in the dynamics of radiocesium in animals, plants, and humans has increased. To elucidate the mechanism of radiocesium transport in animals and plants, visualization of radiocesium in vivo is desirable. Recently we have developed a method for the production and purification of 127Cs. The positron-emitting nuclide 127Cs was produced using the 127I (α, 4n) 127Cs reaction, which was induced by irradiation of sodium iodide with a 4He2+ beam from a cyclotron. We excluded sodium ions by using a material that specifically adsorbs Cs as a purification column and successfully eluted 127Cs by flowing a solution of ammonium sulfate into the column. We injected the purified 127Cs tracer solution into living rats and the dynamics of 127Cs were visualized for 4 hours using positron emission tomography; the distributional images showed the same tendency as the results of previous studies using disruptive methods. Also, we administered the 127Cs tracer to the roots of two species of legume plants; white lupin (Lupinus albus) and soybean (Glycine max) and the dynamics of 127Cs were visualized for 36 hours using a planar positron imaging system. The results showed that 127Cs remained almost completely in the roots of soybean, while 127Cs was rapidly translocated to the aboveground in white lupin, implying that the mechanism of radiocesium transport is completely different even in the same legume plant. These results in this study indicate that the 127Cs tracer is useful for the non-invasive investigation of radiocesium in living animals and plants.

AcknowledgmentThis work was supported by JSPS KAKENHI Grant Numbers JP16K06962 and JP19H04296.
Keywords: Positron emission tomography, Cesium, Environmental impact

Development of New Imaging Method for Visualising Photosynthate Translocation and Release in Plant Root Systems (#426)

Y. - G. Yin1, N. Suzui1, K. Kurita2, Y. Miyoshi1, Y. Unno3, S. Fujimaki4, T. Nakamura5, T. Shinano6, N. Kawachi1

1 National Institutes for Quantum and Radiological Science and Technology, Takasaki Advanced Radiation Research Institute, Takasaki, Japan
2 Japan Atomic Energy Agency, Materials Sciences Research Center, Tokai, Japan
3 Institute for Environmental Sciences, Department of Radioecology, Rokkasho, Japan
4 National Institutes for Quantum and Radiological Science and Technology, Institute for Quantum Life Science, Inage, Japan
5 NARO Hokkaido Agricultural Research Center, Agro-environmental Research Division, Sapporo, Japan
6 Hokkaido University, Research Faculty of Agriculture, Sapporo, Japan


Plant roots release a wide variety of carbon (C) compounds, termed rhizodeposits, into the rhizosphere, which extends several millimeters from the roots into the surrounding soil. These compounds have important effects on microbial populations, nutrient solubility and availability, and can enhance the plant’s ability to cope with adverse soil–chemical conditions. The capability of C partitioning in the root system and the rhizosphere soil differs spatially, therefore, the C apportion process must be regulated by the plant. Realization of the ability to measure and characterize the spatio-temporally changed of C allocation in the root system and the soil, will be valuable for investigation of the regulatory mechanisms of rhizodeposition. In this study, we developed a new method to visualize and evaluate the movement of 11C-photosynthates into the root system and 11C-rhizodeposits released in the soil coupled with positional information using the positron-emitting tracer imaging system. A newly developed rhizobox was consisted of a square nylon mesh bag containing the root system and a pair of soil boxes, and it could be distinguishing the 11C signal of soil from that of root. Test plants of white lupin (Lupinus albus) and soybean (Glycine max) were grown in the rhizobox, and fed 11CO2 as a pulse. Simultaneously, the behavior of 11C-photosynthates into the root system and the distribution of 11C-rhizodeposits released in the rhizosphere soil were visualized. The capability for release of rhizodeposits was characterized from image data for each test plant. White lupin rhizodeposits showed a hotspot of localized high 11C radioactivity, whereas soybean showed uniform spatial distribution of rhizodeposits. The differences were extremely interesting from a physiological point of view, and thus our imaging method can provide accurate information for scientific findings.


This work was supported by JSPS KAKENHI (JP23380155, JP24780251, JP15H04578, JP15H02438, and 19H01169).

Keywords: 11CO2 tracer, Photosynthate, Rhizodeposits, Rhizosphere, Root

Development of projection autoradiography technique using magnetic fields (#650)

K. Kurita1, T. Sakai1, Y. - G. Yin2, N. Suzui2, H. Iikura1, N. Kawachi2

1 Japan Atomic Energy Agency, Materials Sciences Research Center, Tokai, Japan
2 National Institutes for Quantum and Radiological Science and Technology, Takasaki Advanced Radiation Research Institute, Takasaki, Japan


Autoradiography is a useful technique for imaging a distribution of a radioisotope (RI). This technique has good sensitivity and resolution because the sample is required to close contact with an imaging plate (IP) during measurements. In case that there is a distance between the IP and sample, however, the resolution and sensitivity are degraded because beta particles from RI are spread out as a function of the distance. Thus, autoradiography has a difficulty to image samples which cannot close contact to the IP.

Here, we have developed a novel technique named projection autoradiography. The principle is simple. Charged particles in a magnetic field move spirally following the Larmor precession. By utilizing this phenomenon, it is possible to control the trajectory of beta particles emitted from the sample and project the RI distribution onto the IP. Specifically, a uniform magnetic field is applied between the sample and IP, then the beta particles move spirally toward the IP without the severe deterioration of the sensitivity and resolution. In order to prove the new technique, an autoradiography experiment was performed. A radioactive Cs-137 point source (100 kBq, 4 mm in diameter) was prepared as the sample, and neodymium magnets were used. The point source and IP were placed between the magnets. The distances between the point source and IPs were 0, 10, 20, and 30 mm, respectively. The uniform magnetic flux densities were 0, 200, 400, and 650 mT, respectively. The exposure time was 2 min. The autoradiographs showed that the RI distribution in the sample can be projected onto IPs within deterioration of Larmor radius RL, even in the distance of 30 mm between the IP and sample. This result proved that projection autoradiography technique is promising for various applications, such as direct taking a radiograph of cultured samples in a petri dish, etc.

AcknowledgmentThis work was supported by JSPS KAKENHI Grant Numbers JP19K15947.
Keywords: Autoradiography, radio isotope, RI imaging

Initial prototyping of a forceps-type coincidence detector for intraoperative diagnosis of lymph node metastasis (#1163)

S. Ito1, M. Takahashi2, H. G. Kang2, S. Takyu2, K. Kawamura3, F. Nishikido2, Y. Seto4, T. Yamaya2

1 Mirai-imaging corporation, Fukushima, Japan
2 The National Institutes for Quantum and Radiological Science and Technology (QST), Chiba, Japan
3 Chiba University, Center for Frontier Medical Engineering, Chiba, Japan
4 The University of Tokyo, Department of Gastrointestinal Surgery, Tokyo, Japan


18F-FDG is known as a good biomarker for diagnosis of lymph node metastasis, but their intraoperative application has not been studied well. In this work, we designed an intraoperative forceps-type coincidence detector to measure radioactivity concentration in each lymph node. Coincidence pairs of annihilation photons from 18F were measured by using two small scintillators that form the head of the forceps. From initial clinical data, we derived the efficiency requirement of 0.54% or more for the detector. However, the detector head had a size limitation since it had to pass through a typical trocar hole of 12 mm diameter. We made a prototype device by using rectangular BGO crystals with dimensions of 8 mm width x 10 mm length x 4 mm thickness, and we measured sensitivity profiles using a 22Na point source. The sensitivity of 1.18% was obtained for the point source when located at the center of the FOV with the distance of 5 mm between two crystals. However, in the case of a large size lymph node, higher sensitivity will be required because the sensitivity will decrease according to the distance between crystals. Therefore, we followed the experiment by GEANT4 simulation to find an appropriate crystal shape from three candidate shapes: rectangular, half-cylinder and half-shell shapes. We found that the half-shell shape crystal had the highest sensitivity.

AcknowledgmentThis research was supported by AMED under Grant Number JP21hm0102078h0002.
Keywords: PET, Forceps, Monte Carlo simulation

Our exhibitors and sponsors – click on name to visit booth: