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

New York - America ()
Jan 29, 2022, 8:20:29 AM
Your time ()
Tokyo - Asia ()
Jan 29, 2022, 10:20:29 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).

Student Paper Award Competition

Session chair: Hutton , Brian F. (University College London, Institute of Nuclear Medicine, London, UK); Li , Quanzheng (Massachusetts General Hospital & Harvard Medical School, Gordon Center for Medical Imaging, Boston, USA)
Shortcut: M-07
Date: Thursday, 21 October, 2021, 9:15 AM - 11:15 AM
Room: MIC - 1
Session type: MIC Session


Click on an contribution to preview the abstract content.

9:15 AM M-07-01

GPU-accelerated Monte Carlo-Based Scatter and Prompt-Gamma Corrections in PET. (#808)

A. López-Montes1, J. Cabello2, M. Conti2, A. Badal3, J. L. Herraiz1, 4

1 Complutense University of Madrid, Nuclear Physics Group and IPARCOS, Madrid, Spain
2 Siemens Medical Solutions USA, Inc., Knoxville, Tennessee, United States of America
3 US Food and Drug Administration, DIDSR, OSEL, CDRH, Silver Spring, Maryland, United States of America
4 Hospital Clinico San Carlos, Health Research Institute (IdISSC), Madrid, Spain


The accuracy of the quantitative results obtained with Positron Emission Tomography (PET) depends on several factors, being data scatter and background corrections one of the most important ones. Current methods used to estimate the scatter coincidences and the spurious background created by radionuclides that emits prompt γ-rays together with the β+ emission, such as 124I, 86Y, 52Mn, 82Rb, and 44Sc, may be inaccurate in some challenging situations, yielding significant artifacts in the reconstructed images. Monte Carlo (MC) methods can generate accurate estimations of the background present in PET data, but so far, they have been typically too slow for practical applications. In this work, we present the results of the application of the open-source, fast MCGPU-PET simulator within the reconstruction workflow of a clinical PET/CT Biograph Vision scanner. MCGPU-PET provides accurate estimation of true and scatter coincidences and spurious background for non-standard isotopes at around 1 million coincidences per second in 1 single GPU. The results obtained with a bladder and IQ phantom, and patient data show a significant reduction of the artifacts in the final reconstructed images, which in some cases may be up to 100% bias reduction. These results show how state-of-the-art MC methods are already fast enough for clinical applications.


The authors would like to thank Walter Jentzen from Essen (Germany) for the 124I IQ phantom dataset, and John Prior from Lausanne (Switzerland) for the PET brain scan dataset.

Alejandro López Montes and Joaquín L. Herraiz acknowledge support by EU's H2020 under MediNet a Networking Activity of ENSAR-2 (grant agreement 654002), also Comunidad de Madrid (B2017/BMD-3888 PRONTO-CM). European Regional Funds and Spanish Government RTI2018-098868-B-100 (MCIU/AEI,FEDER,EU) RTI2018-095800-A-100 (MCIU/AEI,FEDER,EU) and RTC2019-007112-1

Keywords: Monte Carlo (MC), MC-GPU, Positron Emission Tomography (PET)
9:30 AM M-07-02

Deep Generative Modelling for Enhanced Monte Carlo Simulation of Radionuclide Imaging Data (#507)

J. Moo1, P. K. Marsden1, K. Vyas2, A. J. Reader1

1 King's College London, Biomedical Engineering and Imaging Sciences, London, United Kingdom
2 Lightpoint Medical, Waterside, United Kingdom


Monte Carlo simulations are widely used in radionuclide imaging, including for modelling radionuclide imaging systems, as well as for the development of new image reconstruction algorithms. However, discrepancies in data quality (e.g. spatial resolution) can be observed between simulated and experimental data due to an inability to fully model many bespoke effects, including charge-to-signal conversion processes and detector imperfections. In this study, a deep generative modelling framework is proposed for the enhancement of GATE simulations through the use of cycle-consistent generative adversarial networks (CycleGAN). The networks can be trained in an unsupervised manner to learn the mapping between simulated and experimentally obtained data, obviating the need for paired training data which is difficult to obtain in radionuclide imaging studies. The feasibility of the method was assessed for sensor array outputs from a CMOS intraoperative probe intended for single photon imaging/detection for cancer surgery. Overall, the proposed network was able to learn the distribution of images in both the simulated and the experimental domains. Through the analysis of the Fréchet inception distance (FID) metric, the network was able to achieve a reduction of 94% in the FID score compared to purely GATE simulated data, indicating far greater consistency with measured data.

Keywords: monte carlo simulation, GATE, generative adversarial networks, cycle-consistency, radionuclide imaging
9:45 AM M-07-03

High-sensitivity Cardiac SPECT System Design with Uncollimated Mosaic Detectors: A Simulation Study (#1360)

R. Wang1, 2, Y. Hu1, 2, D. Zhang1, 2, Z. Lyu1, 2, Z. - X. He3, Y. Liu1, 2, T. Ma1, 2

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


Mechanical collimator is the key component in a single-photon emission computed tomography (SPECT) system. It is also the major factor that significantly limits the resolution and sensitivity of SPECT. In this work, we propose to assemble uncollimated mosaic detector blocks with spatially separated scintillators. In this design, one scintillator is naturally collimated by other front-side scintillators. The designed Cardiac SPECT imaging system has 7 scintillator modules surrounding the patient body. Each module consists of 10 scintillator blocks in the axial direction. Each block consists of 89 Χ 89 spatially separated GAGG(Ce) scintillators with a size of 1.68 mm Χ 1.68 mm Χ 20 mm, forming a mosaic pattern. The block is coupled to two SiPM arrays on the top and bottom sides so that the 3-D incident position of a photon is calculated from the dual-end readout signals. When the gamma-ray beam from different directions strikes the detector, the different gamma-event position histograms, i.e. projections, indicate the directions of the gamma-ray. We developed a numerical program to calculate the system response. A numerical hot-rod phantom was utilized to test the image resolution, and a single-slice cardiac phantom was extracted from the XCAT phantom to mimic the cardiac imaging case. The numerical simulation studies demonstrated 6-mm hot rod separation was achievable. Initial Monte Carlo simulations showed that super high detection efficiency (16.71%+/–8.89%) was achievable and the measured projection was prone to the position of gamma sources. The study reveals a promising approach to high sensitivity SPECT imaging. In cardiac imaging, this approach opens a way to a very fast cardiac scan with good resolution. Further works are ongoing to perform Monte Carlo simulations, to further evaluate the image signal-to-noise ratio and impact of body attenuation and scattering, as well as the impact of tracer uptake in other organs.

Keywords: uncollimated system, interspaced mosaic detector, position-sensitive detector, cardiac imaging, SPECT
10:00 AM M-07-04

Improvement of Few-Angle Dedicated Cardiac SPECT Reconstruction using Transformer (#1013)

H. Xie1, Y. - H. Liu2, S. Thorn2, S. Lee3, Z. Liu3, A. Sinusas1, 2, C. Liu1, 3

1 Yale University, Department of Biomedical Engineering, New Haven, Connecticut, United States of America
2 Yale University, Department of Internal Medicine (Cardiology), New Haven, Connecticut, United States of America
3 Yale University, Department of Radiology and Biomedical Imaging, New Haven, Connecticut, United States of America


Convolutional neural networks (CNNs) have been extremely successful in various medical imaging tasks. However, because the size of convolutional kernel used in a CNN is much smaller than the image size, CNN lacks a global understanding of the input images. Transformer, a recently emerged network structure in computer vision, can potentially overcome the limitations of CNNs for image-related tasks. In this work, we proposed a slice-by-slice transformer network (SSTrans-3D) to reconstruct dedicated cardiac single-photon emission computed tomography (SPECT) images from 3D few-view data. To be specific, the network reconstructs the whole 3D volume using a slice-by-slice scheme. By doing so, SSTrans-3D alleviates the memory burden required by 3D reconstructions using transformer. Also, the network can obtain a global understanding of the whole image volume with the attention blocks used in SSTrans-3D. Lastly, already reconstructed slices are used as the input to the network so that SSTrans-3D can potentially obtain more informative features by "seeing" these slices. Validated on pig, phantom, and human studies acquired using a GE Alcyone dedicated cardiac SPECT scanner, the proposed method demonstrates superior performance compared to a deep U-net with conveying paths and multiple dense-net structures. Compared with the deep U-net, cardiac images reconstructed by SSTrans-3D have clearer heart cavity (especially in the apex region), higher heart defect contrast, and overall better quantitative measurements on the testing data.

AcknowledgmentThis work is supported by an internal funding from the Department of Radiology and Biomedical Imaging at Yale University, and the NIH grants R01HL154345 and R01HL123949.
Keywords: Cardiac SPECT/CT, Deep Learning, Nuclear Imaging, Medical Image Reconstruction, Transformer
10:15 AM M-07-05

Image and volume conditioning for respiratory motion synthesis using GANs (#615)

Y. - H. Cao1, V. Jaouen1, V. Bourbonne1, 2, N. Boussion1, 2, U. Schick1, 2, J. Bert1, 2, D. Visvikis1

1 UMR 1101 Inserm LaTIM, Université de Bretagne Occidentale, IMT Atlantique, Brest, France
2 CHRU Brest University Hospital, Brest, France


Four-dimensional computed tomography (4DCT) acquisitions are used routinely in lung cancer radiotherapy treatment planning to identify target volumes and safety margins. However, they expose the patient to higher radiation dose compared to static 3DCT acquisitions. In this work, we demonstrate the possibility of generating synthetic 4DCT acquisitions from a 3DCT image following the actual patient’s respiratory amplitude. To this end, we propose a new image-to-image generative adversarial network (GAN) architecture. More specifically, we propose a new scalar injection mechanism based on Adaptive Instance Normalization to condition the generator on the breathing amplitude. Such information can be obtained in practice using external respiratory tracking devices. We show preliminary results on a series of 4DCT images where we compare our synthesized 4DCT to real respiratory phase-gated acquisitions, paving the way for 4DCT-free treatment planning.

Keywords: Computed tomography, Dynamic imaging, Dose reduction, Artificial intelligence
10:30 AM M-07-06

Voxel-Wise Kinetic Model Selection Using Single-Subject Deep Learning for Total-Body PET Parametric Imaging (#457)

Y. Wang1, 2, E. Berg1, Y. Zuo2, E. Li1, B. A. Spencer1, 2, R. D. Badawi2, 1, S. R. Cherry1, 2, G. Wang2

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


Total-body PET parametric imaging using a single kinetic model may cause inaccurate quantification and artifacts in parametric images because the appropriate tracer model may be tissue and organ dependent. Voxel-wise selection of the best of two or more candidate models can overcome this problem. Conventional methods for model selection are usually time-consuming due to the need for fitting time-activity curves (TACs) to every voxel using every model. In this study, we proposed a deep learning (DL) method for fast voxel-wise kinetic model selection in total-body imaging. The method directly operates adaptively on a single subject and does not require training from a patient population database. A small fraction f of total body voxels are sampled for training the DL model, which is then used to predict the kinetic model to be used for the remaining voxels in the image. The training features are voxel TACs, and the labels are the Akaike Information Criterion (AIC) difference indicating which model should be selected. We have validated the method for a two-model selection problem in four human subjects scanned on the uEXPLORER total-body PET system. The results show that an f of 0.05 reaches a high accuracy (~90%) for voxel-wise model selection while only adding a low computational cost. The DL model selection reduces artifacts in the total-body Ki image and improves image quality as compared to a conventional method without model selection.

Keywords: Kinetic model selection, Kinetic modeling, Total-body PET parametric imaging, Deep learning
10:45 AM M-07-07

Performance Tests Toward Sole Cherenkov TOF-PET (#369)

N. Kratochwil1, 2, S. Gundacker3, G. Terragni1, 4, E. Auffray1

1 CERN, Geneva 23, Genève, Switzerland
2 University of Vienna, Vienna 1010, Wien, Austria
3 RWTH Aachen University, Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, Aachen 52074, North Rhine-Westphalia, Germany
4 University of Milano-Bicocca, Milan 20126, Italy


Time of flight positron emission tomography (TOF-PET) can strongly benefit from a very accurate time estimator, eg. due to prompt Cherenkov emission produced upon 511keV γ-interaction in heavy inorganic scintillators. Considering the quantity sensitivity / (time resolution x cost), crystal solutions based on cheap Cherenkov radiators might outperform standard, Lutetium-based, scintillators due to its low cost and excellent stopping power. Recently it was shown, that coincidence timeresolution (CTR) values as good as 200ps FWHM are possible for 20mm long BGO crystals, despite the low scintillation yield and large decay time.

In fact, scintillating photons are not required to produce an accurate time stamp, opening the door to sole Cherenkov radiators. We evaluated timing performance and sensitivity for PbF2 crystals with various length and surface conditions with Silicon Photomultipliers (SiPMs) and high-frequency readout.

An intrinsic Cherenkov photon yield upon 511keV depositonof 16.5±3.3 photons was measured for 2x2x3mm³ sized crystals. After novel time walk correction based on the slew rate oft he SiPM signal and using all events a CTR of 215ps (159ps) FWHM for Teflon wrapped (black painted) 2x2x20mm³ PbF2 was measured, compared to 261ps (186ps) without correction. For small sized crystals even sub-100ps CTR was measured for 21% of the events (4.3% in coincidence) when selectingon the highest slew rate. Using depth of interaction (DOI) collimated measurements the photon propagation in the crystal was experimentally visualized, resolving the two main waves of photons propagating towards/away from the SiPM.

In this contribution we present performance tests and discuss sensitivty aspects for sole Cherenkov based PET scanners and provide a roadmap addressing current challenges like SiPM dark counts, energy resolution and scalable readout electronics.


This work was performed in the framework of the Crystal Clear Collaboration.

Keywords: Cherenkov radiation, Time-of-flight PET, Lead fluoride, Silicon Photomuliplier
11:00 AM M-07-08

Multi-Isotope Imaging of Therapeutic Alpha Emitters and Their Daughters with an Ultrahigh Energy Resolution Spectral SPECT Imaging System (#1235)

L. Cai1, E. M. Zannoni2, X. Nie1, L. - J. Meng1, 3

1 University of Illinois at Urbana-Champaign, Department of Nuclear, Plasma and Radiological Engineering, Champaign, Illinois, United States of America
2 University of Illinois at Urbana-Champaign, Department of Bioengineering, Champaign, Illinois, United States of America
3 University of Illinois at Urbana-Champaign, Beckman Institute of Advanced Science and Technology, Champaign, Illinois, United States of America


Recently there has been increasing interest in developing a reliable method for quantitative imaging of alpha-emitting radionuclides, which will be helpful to better understanding of radiopharmaceutical uptake and kinetics for radiotherapeutic applications. However, due to the low administered activity and complex decay schemes, a SPECT imaging system for TAT would require detectors with excellent energy resolution capabilities, to discriminate energy peaks from the different radioisotopes involved and to perform simultaneous multi-isotope acquisitions.

Therefore, we have developed an ultrahigh energy resolution spectral SPECT imaging system (referred to as the Alpha-SPECT-mini). It is constructed by 6 detector panels that arranged in a hexagonal stationary gantry. Each detector panel consists of a 2×2 array of detector modules. Each detection unit will offer an excellent spatial resolution (250 μm) and a superior energy resolution (FWHM of 0.80 ± 0.17 keV at 60 keV, and 1.05 ± 0.30 keV at 122 keV).

In the experiment, Ac-225 and Ra-223 filled phantoms were used to evaluate the spectral response of our detection system. It demonstrates the capability of discriminating multi-isotopes such as Ac-225 and its daughters. As for the preliminary imaging study, we found that the projections of capillary tubes are clearly resolved, illustrating a great potential of quantitative imaging even with low activity source.

In this study we will present (a) the Alpha-SPECT-mini system equipped with 24 CdTe imaging spectrometers and a low-noise multi-channel readout circuitry (b) the multi-isotope, multi-functional imaging capability of the system for mapping therapeutic alpha emitter in small animals.

AcknowledgmentWe would like to acknowledge the support from NIH/NIBIB as funding agency (grants R01 EB022388-01 and R01 EB026300-01A1).
We would like to acknowledge the National Institute of Biomedical Imaging and Bioengineering (NIBIB) as funding agency.
Keywords: Targeted Alpha Therapy, SPECT imaging, Multi-isotope imaging, Ultra-high energy resolution

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