PSMR-TBP 2022 9th Conference on PET/MR and SPECT/MR & Total-body PET workshop

28 May 2022, 08:30 → 1 Jun 2022, 04:00 Europe/Rome

La Biodola, Isola d'Elba

Nicola Belcari (Department of Physics "E.Fermi", University of Pisa, Pisa, Italy), Stefaan Vandenberghe (MEDISIP-IBBT-Ugent)

Please go to the PSMR-TBP 2022 official web page for further information.

INFN Privacy Policy

Saturday, 28 May

09:00 → 10:30 Training school on PET/MR and TB-PET reconstruction: Part 1

09:00 Introduction on school

Speaker: Kris Thielemans (University College London)

09:30 Principles of PET and SPECT image reconstruction

Speaker: Kris Thielemans (University College London)

10:00 TotalBody PET challenges

Speaker: Nikos Efthimiou (Technological Educational Institute of Athens)

10:30 → 11:00 Coffee break

11:00 → 13:00 Training school on PET/MR and TB-PET reconstruction: Part 2

11:00 Principles of MR image reconstruction

Speaker: Christoph Kolbitsch (Physikalisch-Technische Bundesanstalt, Braunschweig and Berlin, Germany)

11:30 PyTorch principles for deep-learned iterative PET image reconstruction

Speaker: Andrew Reader (Kings College London)

12:00 Descriptions of projects to be tackled in groups

Speaker: Kris Thielemans (University College London)

12:30 Software set-up

Speaker: Kris Thielemans (University College London)

13:00 → 15:30 Lunch break

15:30 → 17:00 Training school on PET/MR and TB-PET reconstruction: Project work in groups

17:00 → 17:30 Coffee break

17:00 → 19:00 Registration

17:30 → 18:30 Training school on PET/MR and TB-PET reconstruction: Project work in groups

19:30 → 20:30 Welcome cocktail

Sunday, 29 May

08:30 → 08:50 Opening Session: Welcome

Conveners: Nicola Belcari, Stefaan Vandenberghe

08:30 Welcome

TBPET_intro (1).pdf

welcome_talk_rev.pdf

08:50 → 09:30 Opening Session: Keynote speaker

Conveners: Nicola Belcari, Stefaan Vandenberghe

08:50 Shaping the future on nuclear medicine: old and new challenges and the role of emerging technologies

Invited talk.

Speaker: Arturo Chiti (Humanitas University)

Chiti.pptx

09:30 → 10:30 PET/MR and SPECT/MR systems and applications: Part 1

Convener: Nicola Belcari

09:30 The HYPMED PET/MRI Insert for Enhanced Diagnosis of Breast Cancer

Addressing the increasing demands for improved cancer diagnosis and personalized medicine, the H2020 EU project HYPMED aims to develop a PET-RF insert, which is compatible to a 1.5T-MRI Philips Ingenia.
The insert targets the needs of enhanced breast cancer diagnosis, namely offering high-resolution imaging, increased systemsensitivity and also high system-integrity. The PET insert comprises two independent PET rings with a field of view (FOV) of 28 cm × 10 cm, as well as two dual-channel local receive coils. The mechanical designed allows the opening and closing of the PET rings, such that it is possible to perform biopsies without causing major effort and immobilize the breasts. Each PET ring utilizes 14×2 detector stacks comprising three-layered Lutetium–yttrium oxyorthosilicate (LYSO) crystal arrays with 1.3mm pitch. The MR-compatible detectors are based on the Hyperion III platform, using a sensor tile with 12×12 individual digital silicon photomultiplier (SiPM) channels (DPC-3200-22, Philips Digital Photon Counting) having a sensitive area of ∼48mm × 48mm each. The scintillator’s staggered design comprises 3425 crystals and allows the usage of depth of interaction (DOI) information, resulting in high and homogeneous spatial resolution across the FOVs.
We will present the first PET and MRI results of our system.

Speaker: Stephan Niklas Naunheim (Department of Physics of Molecular Imaging Systems (PMI), Institute of Experimental Molecular Imaging (ExMI), RWTH Aachen University)

SN_PSMR_HYPMED.pdf

SN_PSMR_HYPMED.pptx

09:50 Integrated PET/MR scanner as reference imaging tool in the study of dementia: preliminary experience from the PM-D project

The challenge of the screening and follow-up of subjects at risk of dementia is becoming of crucial importance for the sustainability of their assistance by national healthcare. Although imaging biomarkers are playing an essential role for assisting diagnosis and prognosis of dementias, several MR biomarkers are still under evaluation in terms of technical and clinical suitability. Recently introduced simultaneous PET/MR imaging offers a unique opportunity for a comprehensive collection of imaging biomarkers within the same diagnostic examination. In this work we introduce our preliminary experience after the first year of the project PM-D: Integrated PET/MR scanner as reference imaging tool in the study of dementia: technological and clinical assessment, funded by the Italian Ministry of Health. The PM-D project aims to assess the impact of the PET/MR imaging in terms of diagnostic accuracy and benefits/compliance of the patients. PET/MR data, estimated on a cohort of 100 subjects with dementia addressed to perform PET imaging, will be acquired and processed during the project. The project will deliver an imaging protocol that maximizes the trade-off between diagnostic accuracy, patient benefits and cost effectiveness.

Speaker: Marco Aiello (IRCCS SDN, Naples)

Aiello_psmr2022_OK.pptx

10:10 A dedicated MRI/PET System for Radiotherapy Treatment Simulation

The Unity 1.5T MR-linac has clinically been introduced to provide stereotactic precision for moving targets. We are now developing an integrated 1.5T PET/MRI system with identical MRI performance for planning of MR-linac treatments. The system offers 70 cm bore size and includes PET detectors mounted in the gap of a split gradient coil using the Hyperion III platform. Recently, the first ring of PET detectors was installed. We performed extensive interference tests and acquired the first simultaneous images using a brain phantom. Our vision is that the combination of MR-linac with PET/MRI will allow the use of PET information for stereotactic radiation therapy guidance.

Speaker: Dr David Schug (RWTH Aachen University, PMI - Physics of Molecular Imaging; Hyperion Hybrid Imaging Systems GmbH)

2022-05-28_PSMR2022_Schug_UMCU_PETMR_Radiotherapy_Planing_noAppendix.pptx

10:30 → 11:00 Coffee break

11:00 → 12:40 PET/MR and SPECT/MR systems and applications: Part 2

Convener: Marco Carminati

11:00 Design of the BrainPET 7T Project’s UHF MRI-Compatible PET Modules based on the Hyperion III Platform

PET-MRI provides excellent soft-tissue contrast combined with visualization of metabolic processes. Especially, brain-imaging benefits from organ-dedicated scanners providing high sensitivity and homogenous spatial resolution. The BrainPET 7T project aims to build a PET insert for human imaging with a particular emphasis on UHF MR-compatibility, imaging of non-proton nuclei, and neuroscientific studies with cognitive tasks. The system design enables a sensitivity of 12% and a spatial resolution of 1.6 mm based on point source and Derenzo phantom simulations. Here, we present the concept and implementation of the MRI-compatible PET modules used in the BrainPET insert.
The BrainPET insert contains 8 identical PET modules which employ the Hyperion III PET platform as readout electronics. 15 detector stacks with a 48 x 48 mm$^2$ active area are equipped with digital SiPMs (DPC 3200-22, DPC) and connected to one singles processing unit (SPU). The synchronized SPUs collect, (pre-)process, and send the data to a central data acquisition architecture.
For every PET module, three highly-integrated Mounting & Cooling (M&C) structures host five detectors each, resulting in an axial FOV of approximately 25 cm. The M&C structure is a composite consisting of glass-reinforced plastic for mechanical stability and a copper inlay featuring liquid cooling. As the copper inlay is not required for mechanical stability, we were able to highly structure the electrically conductive area to reduce Eddy currents induced by the MRI’s gradient switching. All detector stacks connect to the M&C structure utilizing a kinematic mounting system, ensuring a precise positioning while preventing strain energy to be inserted into the system.
A carbon-fiber-reinforced plastic housing encloses all components reducing interferences between MRI scanner and PET electronics. A sizeable removable cover of the housing allows installation and maintenance of all components.

Speaker: Dr David Schug (1 Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, RWTH Aachen University, Aachen, Germany 2 Hyperion Hybrid Imaging Systems GmbH, Aachen, Germany)

2022-05-28_PSMR2022_Schug_HVF_BrainPET_PET_Modules_noAppendix.pptx

11:20 Preliminary Study of SPECT-MRI Compatibility for a Clinical SPECT-MRI INSERT

The INSERT (INtegrated SPECT/MRI for Enhanced stratification of brain tumors in Radio-chemoTherapy) is the world's first clinical SPECT-MRI brain imaging system. Here we demonstrate its use within a clinical MRI environment for the first time. The INSERT has been evaluated as a stand-alone SPECT scanner and preclinical designs demonstrate the feasibility of simultaneous SPECT-MRI. Customised MRI sequences are implemented here to overcome system interference and field inhomogeneity introduced by presence of the INSERT in the MRI during simultaneous acquisition. We also acquired SPECT data within the MRI environment to evaluate the SPECT system performance. Finally we present a set of sequential SPECT-MRI data, demonstrating imaging capabilities of the complete system.

Speaker: Mr Ashley Morahan (University College London)

08_Morahan.pptx

11:40 Integrating an annulus-shaped transmission imaging source into the body coil of a PET/MRI system: influence on MR imaging

Accurate correction for photon attenuation is essential for quantitative simultaneous positron emission tomography/magnetic resonance imaging (PET/MRI). Several methods have been developed to derive attenuation maps based on MRI or emission data, but they all have limitations affecting their robustness and general applicability, especially in whole-body PET/MRI. As an alternative, a transmission imaging system based on an annulus-shaped source was proposed earlier. For this system, a prototype was implemented as a removable insert and it was successfully used to generate attenuation maps. In this study, we aimed to test the feasibility of integrating an annulus-shaped transmission source into the body coil of a simultaneous PET/MRI system. We showed that installing a liquid-filled hose between the RF shield and the rungs of the body coil does not significantly affect the MRI signal-to-noise ratio, and that it does not generate signals disturbing the MRI process when excited by its RF pulses (no fold-in artifacts were observed). We conclude that an annulus-shaped transmission source can be integrated into the body coil of a whole-body simultaneous PET/MRI system without compromising MRI performance. Installing the annulus at this location has a big advantage: it allows a robust method of obtaining accurate attenuation maps without sacrificing space in the patient bore. This will be of great benefit in particular for the development of wide-bore systems, which is necessary to enable simultaneous PET/MRI for a wider range of patients and applications such as radiotherapy planning.

Speaker: Woutjan Branderhorst (UMC Utrecht)

2022-05-29 PSMR 2022 presentation PET-MRI-TRANSMISSION.pptx

12:00 Dead Time Effects and Image Quality Evaluation at High Activities for the SAFIR Dual Ring Prototype

The SAFIR Dual Ring Prototype is a high-performance preclinical PET-Insert designed to operate at activities of up to 500 MBq within the bore of a 7T Bruker BioSpin MRI system.
It features an axial coverage of 35.6 mm with a coincidence resolving time of 194 ps FWHM, an energy resolution of 13.8% FWHM and has been fully characterized according to the 2008 NEMA NU-4 standard.

The focus of this study was to examine the detector performance at high activities with respect to the image quality following the NEMA NU-4 standard.
We evaluated the dependency of the spill-over ratio, recovery coefficient and uniformity on the total activity within the phantom for up to 500 MBq.
Furthermore, we investigated the effects of the detector dead time per channel at different activities.

A maximum relative increase in the spill-over ratio of 7.3% for water and 5.1% for air was observed after applying all corrections, while no noticeable trend was observed for the recovery coefficients.
The percent standard deviation of the phantoms uniform region improved by 8.8%.
At 484 MBq, 11% of single events were lost due to dead time effects.

Lastly, we present a method to correct PET image data for the effects of detector dead time, and showcase the effects of said method on the image quality.

Speaker: Jan Debus (ETH Zürich)

PSRM_2022_SAFIR_JanDebus.pdf

12:20 3D Printed Alignment Apparatus for Retrofitted Rodent PET-MRI at 9.4 Tesla

We propose a retrofitted alignment apparatus that attaches to the MRI’s motor-driven automated positioning system and enables simultaneous PET-MRI at 9.4T with a stand-alone microPET ring. The system leverages 3D printing technology to achieve stable alignment between the two modalities, consistent placement of animals, and rapid reconfiguration between PET-MRI and MRI-only experiments.

Speaker: Mr Leo Marecki (Buffalo Neuroimaging Analysis Center, Department of Neurology, Jacobs School of Medicine and Biomedical Sciences at the University at Buffalo, Buffalo, NY, USA)

Leo Marecki PSMR 2022 Oral Presentation.pptx

12:40 → 15:30 Lunch break

15:30 → 16:50 Total Body PET clinical systems and applications

Convener: Kuangyu Shi

15:30 Invited talk - Long axial FOV PET hands-on: first view and future perspective for Emi/Total-Body PET

Speaker: Riemer Slart

TBPET PSMR 2022 Riemer Slart-2.pptx

16:10 Evaluation of population-based input functions for kinetic modelling of 18F-FDG datasets from a long axial FOV PET scanner

Accurate estimation of the available tracer concentration in the plasma, known as the arterial input function (AIF), is essential for kinetic modelling of dynamic PET datasets. The gold standard method to measure the AIF requires collection of serial arterial samples. With the introduction of long axial field of view (LAFOV) PET system, image derived input functions (IDIF) can be reliably measured from large blood pools such as left ventricle and aorta. However, measurement of an IDIF still requires a long dynamic PET acquisition which can be unpractical in a clinical setting. In this work, we exploit the high sensitivity and temporal resolution of LAFOV PET systems and study the feasibility of PBIFs with abbreviated protocols in 18F-FDG total body kinetic modelling. Dynamic PET data were acquired from 24 oncological subjects for 65 minutes following the administration of 18F-FDG. The data were split into 16 training and 8 testing sets, and IDIFs were extracted from the descending aorta. The training datasets were used to generate a PBIF. We compared use of different scan durations for generation of scaled PBIFs (sPBIF) and performance of different Patlak start time, t, in Ki estimates. The sPBIF55-65 demostrated the best performance with 1.5% bias and 6.8% precision. Using the sPBIF55-65 with 20 minutes of PET data (t=45) was adequate to achieve <15% precision error on Ki estimates of tumour lesions compared to Ki estimates with IDIF. In brain grey matter, sPBIF55-65 with 15 minute of PET data (t*=50) yielded Ki estimates with less than 1% bias and less than 15% precision error. The use of PBIFs with shortened protocols can enable wider adoption of parametric imaging protocols in clinic setting.

Speaker: Hasan Sari

PSMR_Sari_PBIF_May29.pptx

16:30 Quantitative Image Derived Input Function from Long Axial Field of View scanners

Introduction: With the recent introduction of Long Axial Field of View (LAFOV) clinical scanners with simultaneous acquisition of a field of view (FOV) of 1m or more, the sensitivity of PET scanners compared to otherwise state-of-the-art PET scanners with FOVs of 20-25cm, has increased dramatically due to the increased angular acceptance of coincidence events in the scanner. This creates new possibilities for input derived image functions (IDIFs) with high time resolution, low noise, and high spatial resolution.
Here we present the automatization of an IDIF generation on a high sensitivity Siemens Quadra scanner using [15O]-H2O. The derived IDIF is compared to arterial sampling and organ specific delays are estimated based on an automatic segmentation to illustrate the unique possibility of a self-consistent scan including organ specific input functions for whole-body perfusion data in a single session.
Methods: 5 clinically referred patients were included in the study after written consent. 400 +/- 50 MBq [15O]-H2O was injected for each scan session. Each patient underwent either one or two rest and acetazolamide (Diamox) phases, respectively, totaling 16 scan sessions. In one patient arterial blood was sampled by an Allogg automatic blood sampler with a 1 sec. sampling rate matching the shortest PET time framing.
PET data was acquired for a minimum of 3 min. in each session and subsequently reframed in 1x5s, 30x1s, 15x2s, 5x10s, 3x20s frames and were reconstructed using 3D-OP-OSEM with 4 iterations and 5 subsets including time of flight information in a 440x440x645 matrix resulting in 1.65x1.65x1.65 mm3 voxels.
The aorta was segmented by the SegTHOR algorithm on corresponding CT scans. The mask defined by SegTHOR was continuously pixel-wise eroded and optimized with regards to the area under the input function curve (AUC) and keeping the pixel noise in the curve low.
Organ segmentation including liver, lungs, kidneys, spleen, and bone were segmented using a deep learning network on the CT scan as implemented in the MIWBAS research prototype. The extracted masks where hereafter used to extract the mean tissue time activity curves.
Delays from both arterial input sampling and IDIFs were delay fitted using tpcclib (Turku PET centre, version 0.7.6) to the respective regions defined by the masks from MIWBAS. Before fitting, the masks were eroded by two voxels to minimize any edge artifacts on the statistical parameters of the volume. Dispersion was not considered for the IDIFs.
Results: The upper descending part of aorta was selected as a reference segment as the best comprise between diameter of the segment hence minimizing partial volume effects, low spill-in and the relatively small degree of motion in this segment.
An erosion corresponding to a final volume of approximately 4 mL was found to give both the best AUC agreement between arterial sampling and IDIF for both rest and Diamox phase. A good general agreement of peak height was found for the selected degree of erosion between delay and dispersion corrected AIF and IDIF. Additionally, the AUC and peak height were seen to converge towards a stable value indicating minor influence of partial volume effects after this point. The fitted mean delays for right kidney, left kidney, brain, liver (v. portae) and spleen were 3.5s, 2.7s, 2.3s, 18.4s and 1.4s, respectively, agreeing well with literature values.
Conclusion: An automatically derived aorta IDIF from convolutional neural network-based segmentations has been evaluated and compared to an arterial input function. We demonstrate the feasibility of an automatic input function segmentation pipeline on a LAFOV scanner allowing an individual organ input function hence minimizing kinetic model parameter estimation. This enables simultaneous whole-body perfusion in a single scan session with reliable IDIF as verified by comparison to arterial sampling.

Speaker: Thomas Lund Andersen (Department of Clinical Physiology, Nuclear Medicine and PET, Rigshospitalet, University of Copenhagen, Denmark)

IDIF_TLA_PSMR.pptx

16:50 → 17:20 Coffee break

17:20 → 19:30 Special session on PET/MR and TB-PET phantoms

Conveners: Kris Thielemans, Stefaan Vandenberghe

Connect to the session

17:20 Need for PET/MR and TB-PET phantom standardization

Speaker: Charalampos Tsoumpas (University Medical Center Groningen)

Tsoumpas.pptx

17:40 Quantitative myocardial perfusion imaging. A novel multimodality phantom validation

Speaker: Riemer Slart

2022 Riemer Slart en Marije Kamphuis heart phantom.pptx

17:55 3D printing of an anthropomorphic head phantom for PET/MRI

Introduction: Quality assurance tests of functional nuclear medicine imaging with positron emission tomography (PET) in combination with MRI requires standardized phantoms visible in both modalities. The required measurements are usually performed using homogeneously filled PMMA phantoms. However, anthropomorphic, heterogeneous phantoms are needed to replicate patient examinations in the best possible way and consider attenuation correction correctly. For this purpose, a realistic human skull was reproducibly manufactured by 3D printing using alabaster casting techniques.

Methods: Based on the anatomical bone structures from MIDA model [1] a sagittal bisected negative mold was calculated and transform into a printable file. 3D printing was performed using water-soluble polyvinyl alcohol (PVA). Following some casting preparation, both printed molds were cast with alabaster-mixture. This material imitates bone and provides properties comparable to human cortical bone. After the curing process, the halves were put in water so that PVA dissolved and the gypsum enriched with water. This step is decisive for the phantom to give a signal in MRI. Finally the gypsum halves were composed and filled up with superabsorbent for brain imitation, in which a PVC flask was placed for activity filling as already shown by Harries et al. [2].

Results: The created phantom was used for cross-calibration and measured using UTE and DIXON sequence in PET/MRI. Both sequences obtained a realistic attenuation map segmentation. Based on this segmentation, the attenuation coefficients could be assigned correctly and the reconstructed and attenuation corrected images represent the expected activity concentration and linear attenuation coefficients. Compared to an activity standard (filled spherical phantom) the accuracy of the PET measured activity concentration was 0.96 for DIXON- and 0.90 for UTE- based attenuation correction.

Conclusion: It was shown that structures of the human skull can be realistically and reproducibly imitated by 3D printing a segmented negative mold and subsequent alabaster casting. Furthermore, the created phantom could be segmented correctly by means of MRI.

References
[1] Iacono MI, Neufeld E, Akinnagbe E, Bower K, Wolf J, Vogiatzis Oikonomidis I, Sharma D, Lloyd B, Wilm BJ, Wyss M, Pruessmann KP, Jakab A, Makris N, Cohen ED, Kuster N, Kainz W, Angelone LM (2015) MIDA: A Multimodal Imaging-Based Detailed Anatomical Model of the Human Head and Neck. PLoS One, 10[4], e0124126, doi:10.1371/journal.pone.0124126, 22.04.2015, PMID:25901747

[2] Harries J, Jochimsen TH, Scholz T, Schlender T, Barthel H, Sabri O, Sattler B (2020) A realistic phantom of the human head for PET-MRI. EJNMMI Phys, 7[1], 52, doi:10.1186/s40658-020-00320-z, PMID:32757099

Speaker: Mr Benedikt Bothe (University Medical Centre Rostock, University Medical Centre Leipzig)

PSMR22_Presentation_BB.pptx

18:10 Physical MRI phantoms from radiomics perspective

MRI has a number of distinct clinical advantages: e.g. multi-sequence capabilities producing superior contrast among soft tissues, full 3D imaging, and no ionizing radiation. Additionally, quantitative imaging biomarkers (referred to as radiomics) based upon MRI show promising single-study results for diagnosis, treatment monitoring, and outcome prediction. While relatively simple measures (shape and first order statistics based upon histogram) do provide some merit, full scope of radiomics is utilised by addition of heterogeneity analysis. That is specially the case for MRI where clinical imaging often uses relative pixel values (T1-weighted or T2-weighted) instead of, more complicated and time consuming options providing quantitative data.
However, radiomics still lack reproducibility originating in absence of standardisation in terms of different manufacturers and different approaches to analyse relatively complicated tissue heterogeneity. A part of the solution is in physical phantoms containing sufficient information that would allow grasping the textural differences between tissue types or processes of interest. These phantoms will be used for optimization of radiomics pipeline consisting of a variety of technical solutions for interpolation, discretisation, feature extraction, feature selection, and machine learning.

Speaker: Dr Lejla Alic

PSMR-TBP22 --- Lejla Alic ---FINAL.pptx

18:25 Image Quality Phantom Comparison of the Biograph Vision Quadra and Biograph Vision PET/CT Scanners

Comparing scanners performance can be a rather complicated task, since the NEMA protocol was not designed considering long axial field of view PET scanners. This work compares several image quality measures using the NEMA IQ phantom for the Biograph Vision and Biograph Vision Quadra PET/CT scanners. Two acceptance angles of the Vision Quadra were considered in the comparison. The aim was to quantify the improvement introduced by the Vision Quadra compared to Vision in terms of noise without degrading spatial resolution. Results showed that the higher sensitivity of Vision Quadra allowed to reduce the activity by 3.9$\times$ or collect the same amount of trues 8.1$\times$ faster compared to Vision, while the contrast recovery coefficient showed small differences between both scanners.

Speaker: Hasan Sari (Advanced Clinical Imaging Technology Siemens Healthcare AG)

psmr22_quadra_Phantom_29May.pptx

18:40 Phantom Studies to Characterize Total-Body PET System Performance

Introduction: Total-body PET systems have two main advantages over scanners with standard axial fields-of-view (AFOVs): their high sensitivity and the capability to image dynamic processes in major organs of the body simultaneously. NEMA PET performance standards (NEMA NU-2) were developed more than 2 decades ago for scanners with AFOV of 65 cm or less, and these measurements may not adequately reflect the performance differences of TB-PET systems compared to those with standard AFOV. Thus, it may be necessary to modify the measurements to handle both TB-PET and standard PET systems, while recognizing that any change should be driven by an interest in reflecting performance of all scanners in the clinic. Methods: We have performed standard NEMA tests on the PennPET Explorer, a system with 142-cm AFOV, augmenting them with additional tests that highlight the behavior of a TB-PET system. For the sensitivity measurement, while the standard 70-cm line source is indicative of counts that would be detected in the range of the most common whole-body clinical surveys (i.e., eyes-to-thighs), an additional measurement with a 170-cm line, which is the average height of an adult human, better reflects the full advantage of a TB-PET system with increased acceptance of oblique coincidence pairs. An even longer source would provide useful information about the variation in sensitivity across the AFOV, particularly for a system such as the uEXPLORER with 194-cm AFOV. In the measurement of count rate performance, because injected activity does not leave the AFOV of a TB-PET system during a dynamic study (except through physical decay), a 70-cm phantom underestimates the randoms fraction, and overestimates the noise equivalent count rate seen in the clinic. For this reason, we also have imaged two 70-cm phantoms back-to-back to form a 140-cm long distribution, more reflective of the range where activity accumulates in the body. The axial spatial resolution can degrade in the center of the AFOV of a TB-PET system due to depth-of-interaction uncertainties for oblique lines of response; therefore, a measurement of spatial resolution at more than two axial locations characterizes any spatial variation more fully. In addition, the use of rebinning techniques prior to the prescribed filtered back-projection reconstruction can alter the point spread function in a way that is not observed in images reconstructed with the clinical algorithms - notably, iterative reconstruction. The requirement for an analytic algorithm does not reflect performance of the system under clinical conditions and should be reassessed, being mindful of the potential for iterative reconstruction of point sources in air to result in artificially better (lower) spatial resolution. Similarly, there is merit in acquiring the image quality phantom measurement at several axial locations to demonstrate any impact of spatial resolution and sensitivity variations with axial position on contrast recovery and image noise. Finally, we have used a long cylindrical pipe phantom that extends through the AFOV to assess the uniformity of image noise and quantitative accuracy throughout the AFOV, since the axial sensitivity and correction factors (e.g., attenuation, detector normalization) vary quite dramatically from the center to the end of the system. Conclusions: In order to properly compare systems and estimate performance in practice, standard measurements of performance may need to be adjusted to handle TB-PET systems, either by modifying the activity distribution or the processing/analysis. We have performed the NEMA measurements and additional studies on the PennPET Explorer, a scalable TB-PET system that has operated with an AFOV ranging from 64 cm to 142 cm, and have gained understanding of how phantom studies relate to performance of human studies with a variety of radiotracers. The goal is to characterize the system in a way that predicts performance, with measurements that apply to all PET systems, and provide guidance for future NEMA NU-2 updates.

Speaker: Margaret Daube-Witherspoon (University of Pennsylvania)

PSMR_2022_TB_tests_v3.pptx

18:55 Round table

Monday, 30 May

08:30 → 10:50 Metabolic Imaging with PET, MR and PET/MR

Conveners: Michela Tosetti, Paola Erba

08:30 Imaging Gas-Exchange Lung Function using Density-Weighted MRSI and Hyperpolarised 129Xe Gas

3D density-weighted MRSI in combination with a frequency-tailored RF excitation pulse was designed, implemented and used to detect xenon gas in the lungs and xenon dissolved in lung tissue and blood. These images were used to calculate quantitative ratio maps of tissue-to-gas, blood-to-gas, and blood-to-tissue with good SNR.

Speaker: Rolf Schulte (GE Healthcare)

Schulte_20220530_rfs_psmr_xenon.pptx

08:50 Non-invasive PET/MR imaging of CBF and CMRO2

The gold standard for imaging cerebral blood flow (CBF) and the cerebral metabolic rate of oxygen (CMRO2) are complex and invasive PET techniques. Hybrid PET/MR imaging can greatly simplify the procedures by acquiring PET data while simultaneously obtaining MRI measurements of whole-brain (WB) CBF and CMRO2. The aim of this work is to present hybrid PET/MR methods to image CBF and CMRO2. PET/MR imaging of CBF (PMRFlow) and oxidative metabolism (PMROx) can either calibrate PET data explicitly by incorporating the WB MRI measurements into the PET kinetic model, or by scaling the IDIF obtained from the WB time-activity curve and performing a fitting routine. The implementation of PMRFlow and PMROx allow for head-to-head comparisons between emerging MR methods and well-established PET measurements for imaging cerebral metabolism.

Speaker: Dr Keith St Lawrence (Lawson Health Research Institute)

PSMR2022_Narciso_2022.pptx

09:10 Cerebral Blood Flow assessed by MRI in a naturalistic cohort of MCI subjects.

FDG-PET has been proven so far to be a good modality for detecting functional brain changes in AD, for identifying changes in early AD, and in helping to differentiate AD from other causes of dementia. However, in recent years, novel MRI techniques, such as pCASL, showed to be promising in investigating brain metabolism and neurodegeneration.
In this work, a multiparametric approach is proposed to investigate the structure and the metabolism of the brain in a group of subjects with Mild Cognitive Impairment in order to verify whether Cerebral Blood Flow (CBF) measured by MRI can be used as a quantitative neuroimaging biomarker, alternative to FDG-PET, to differentiate disorders profiles and to assess quantitively the outcome in rehabilitative interventions.
A cohort of 78 subjects participated in the MRI study: 10 healthy elderly (HC, 8 females, F, mean age ± standard deviation = 73±4y) and 68 MCI subjects (32F, 76±5y). A half of the MCI subjects followed their natural lifestyle, while the other half underwent a multi-domain training for 7 months. The protocol provided to perform an MRI exam at T0 and a follow-up exam after 7 months (T7).
A Voxel Based Morphometry approach show to be able to detect CBF alterations of MCI subjects with respect to healthy controls at time zero and can pinpoint regional CBF increase after 7 months in the trained subjects. Moreover, an unsupervised clustering method proved to be effective in identifying four different structural neurodegeneration classes (on the basis of morphometry measurements derived from T1w structural imaging) which are associated also to different metabolic profiles as assessed by MRI-based CBF measurements.
This multiparametric approach shows to be promising in differentiating MCI subgroups and in monitoring MCI progress along time, as in the natural history of the disease as during pharmacological and non-pharmacological treatments.

Speaker: Laura Biagi (IRCCS Fondazione Stella Maris, Pisa, Italy)

03_Biagi.pptx

09:30 Spatiotemporal learning of dynamic positron emission tomography data improves diagnostic accuracy in breast cancer

Abstract
Positron emission tomography (PET) can reveal metabolic activity in a voxel-wise manner. PET analysis is commonly performed statically by analyzing the standardized uptake value (SUV). A dynamic PET acquisition can provide a map of the spatio-temporal concentration of the tracer in vivo, hence conveying information about radiotracer delivery to tissue, its interaction with the target and washout. Tissue-specific biochemical properties are embedded in the shape of time activity curves (TACs), which are conventionally employed along with information about blood plasma activity concentration i.e., the arterial input function (AIF), and specific compartmental models to obtain a full quantitative analysis of PET data. The main drawback of this approach is the need for invasive procedures requiring arterial blood sample collection during the whole scan. In this paper, we address the challenge of improving the PET diagnostic accuracy through an alternative approach based on the analysis of time signal intensity patterns. Specifically, we demonstrate the diagnostic potential of tissue TACs provided by a dynamic PET acquisition using various deep learning models. Our framework is shown to outperform the discriminative potential of classical SUV analysis, hence paving the way for a more accurate PET-based lesion discrimination without additional acquisition time or invasive procedures.

1. Introduction
Positron emission tomography (PET) allows the quantification of the biochemical properties of the tissue under investigation through the injection and detection of a targeted radiotracer [1]. To date, PET tissue time activity curves (TACs), and dynamic PET data in general, are mainly used for kinetic modelling, which involves fitting tracer-specific compartmental models[4]–[7] and requires information about blood plasma activity concentration, i.e., the so-called arterial input function (AIF), which requires arterial cannulation and collection of blood samples during the whole PET acquisition (which can last up to 90 minutes) to be measured. In rare cases, the shape of the TACs is evaluated visually to qualitatively discern tissue types.
Currently, the standardized uptake value (SUV), or its normalized version (SUVR), are the most widely employed PET-derived measures both in clinical and research application. These static maps are equivalent of the late phase of a dynamic PET acquisition, and therefore discard most of the information possibly present in the time-evolution of tissue-specific TACs. Importantly, the quantitative accuracy of SUV estimates relies on the many assumptions[8] which fail in clinical PET, resulting in non-negligible errors in the estimation of the rate of tracer uptake. PET tracer distribution is a dynamic process influenced by a wide number of factors (e.g., tissue type, patient, time of scan) which are reflected in temporal PET dynamics and cannot be accurately accounted by static SUV imaging.
The aim of this paper is to explore the information content and added value of employing tissue TACs only (i.e. eliminating the need for arterial blood sampling). To this end, we combine several deep learning architectures to analyze clinical data obtained from a cohort of breast cancer patients who received dynamic 3′-deoxy-3′-18F-fluorothymidine (18F-FLT) PET scans.

2. Methods

A. Dataset
We employed a publicly available clinical 3′-deoxy-3′-18F-fluorothymidine (18F-FLT) dynamic PET data set of 44 breast cancer patients, part of the "ACRIN-FLT-Breast (ACRIN 6688)" collection in the Cancer Imaging Archive (TCIA)[10]–[12].

B. Data pre-processing
For each patient, consecutive regions of interest were manually contoured around the tumor by an experienced radiologist on the static PET image (obtained as the average of the last 5 timeframes of the dynamic PET data). The 18F-FLT radioactivity concentrations within the volumes of interests were normalized to injected radioactivity and body weight to obtain SUV values[8]. For each patient, the masks obtained from lesion segmentation were flipped on to the contralateral breast for the delineation of a reference healthy region. For each patient, a median of 574 (range, 63 – 6954, according to lesion size) TACs were extracted from using the above-mentioned reference and lesion masks. TACs were linearly resampled onto a uniform time axis (one sample every 10 seconds for a total of 331 samples) (Fig. 1).

C. Spatiotemporal models for dynamic PET data
We compare convolutional mono-dimensional models (CONV1D) to long short-term memory (LSTM) models, and also evaluate the performance of a combined model (CONV1D+LSTM). For comparison to standard PET analysis, voxelwise SUV data were extracted from both lesion and reference tissue and used as input for an RUS boost tree ensemble classifier [14] as well as a boosted tree technique and a support vector machine. For each model, hyperparameter optimization was performed in the optima library (with random search sampler; 200 trials) and involved the number of units (for fully connected and LSTM layers), the number of filters and the dimension of the stride (for convolutional layers), the activation function, the learning rate, the loss function, the metric, gamma, C and kernel for the support vector machine (SVM) classifier and, for the XGboost, the maximum depth of a tree, number of estimators, and the fraction of columns to be subsampled. In the following sections, optimized values are listed.
CONV1D: Two mono-dimensional convolutional layers using a rectified linear unit (relu) and linear activation function, respectively. The filter size was set to 16 and 32, the kernel size to 2 for the first convolutional layer and to 4 for the second one and the stride to 2 and 3. The output of the last convolutional layer was flattened into for four fully connected layers with 64 neurons followed by the last softmax layer for classification.

LSTM: Two LSTM and two fully connected layers. The units of the LSTM layers, using a sigmoid activation function, were set to 16. The output of the last LSTM layer was flatted into two fully connected layer with 64 neurons followed by the last softmax layer for classification.

CONV1D+LSTM: For spatiotemporal feature extraction, the model included both convolutional and recurrent neural networks. This architecture combines the previous two.

TRANSFORMER: The transformer model consisted of stacked self-attention and pointwise, fully connected layers for both the encoder and decoder. It was adapted for timeseries classification by Vaswani et al. For full details, see [15].

D. Implementation
All experiments were conducted using python version 3.8, the keras deep learning library using TensorFlow as backend. We employed a Linux machine and two Nvidia Pascal TITAN X graphics cards with 12 GB RAM each.

E. Performance and evaluation
The sample was split into training (80%), validation (10%) and testing sets (10%). An early stopping method was used to select the optimum number of training epochs and the batch size (Keras callback function monitoring the loss function with a patience set to 10). Models were trained for 200 epochs on a batch size of 64 and evaluated on the independent test set [17].

3. Results
Table 1 summarizes the results obtained with the validation of our models in terms of accuracy, precision and recall. When classifying 1D time series, the best performance was obtained by the CONV1D model with a 87% accuracy in comparison to 81% accuracy obtained with the LSTM and a 77% accuracy obtained with a combination of the two (CONV1D+LSTM). Transformer models classified lesion TACs with 69% accuracy. In comparison to SVM and XGboost used on the most commonly employed SUV imaging technique, which discriminated tumour tissue with 84% and 50% accuracy, respectively, our CONV1D model provided more stable results (confusion matrix – Table 1).

4. Discussion and Conclusion
We employed mono-dimensional filters to learn temporal patterns from time sequences data. The performance of our models was finally compared to the gold-standard SUV method. This proof-of-concept study demonstrated that the diagnostic accuracy of a static PET can be easily improved with a non-invasive deep learning approach which exploits the biochemical and metabolic information embedded in the tissue time activity curves obtained with a dynamic PET acquisition. Our results pave the way for more specific and sophisticated applications where, deep-learned time signal intensity pattern analysis can be used for tumor segmentation or, more interestingly, for tracer kinetic assessment without any pharmacokinetic model or measurement of the AIF.

References
[1]Gupta, Chest 1998; [2] Thorwarth, BMC Cancer 2005; [3] Sinibaldi, J Tissue Eng Regen Med 2018; [4] Sharma, Eur J Nucl Med Mol Imaging 2020; [5] Sharma, J Nucl Med 2020; [6] Dubash, Theranostics 2020; [7] Li, Pharmaceutics 2021; [8] Westerterp, Eur J Nucl Med Mol Imaging 2007; [9] Karakatsanis, Phys. Med. Biol 2013; [10] Kinahan, http://doi.org/10.7937/K9/TCIA.2017.ol20zmxg; [11] Kostakoglu, J Nucl Med 2015; [12] Clark, J Digit Imaging 2013; [13] Conti, Seminars in Cancer Biology 2021; [14] Seiffert, Trans. Syst., Man, Cybern 2010; [15] Vaswani, http://arxiv.org/abs/1706.03762; [16] Zhang, http://arxiv.org/abs/1708.06578; [17] Šimundić, EJIFCC 2009

Speaker: Marianna Inglese (Department of Biomedicine and Prevention, University of Rome Tor Vergata)

PSMR22-IngleseM.pptx

09:50 Forecasting the pharmacokinetics in dynamic brain PET imaging using Neural Ordinary Differential Equation: a step forward for dual-tracer studies

In dynamic brain positron emission tomography (PET) studies, recovering the images in the missing time frame is often required in order to reduce the scanning protocol, or to perform kinetic modelling with sparse dynamic information. Likewise, the rapid dual-tracer studies, which aim to administer two tracers in a single scan staggered in time, will largely be benefitted if the later frames of the first administered tracer can be predicted where both tracers overlap, as it will allow to single out the signal from the second tracer as well by simple subtraction from the total measurement. This indicates that each individual tracer information would be recovered, which has been a challenge owing to the fact that all tracers give rise to indistinguishable 511KeV annihilation photon pairs as a signal to PET scanner. Traditionally, this was done by setting a kinetic model that consists of sets of ordinary differential equations (ODE), such as parallel compartment model, and fitting the measured time-activity curve (TAC). Here, we introduce the novel deep learning model, neural ODE which shares the same concept in data-driven manner. Simply put, the neural ODE solves sets of ODE and converges into the functional shapes that best describe the underlying pharmacokinetic processes. We customized the neural ODE and applied the proposed model to 60-minute dynamic 18F-PI-2620 brain PET images such that it will predict the late 30-min kinetics, given the early 30-min frame images as an input.

Speaker: Jimin Hong

PSMR_Hong.pptx

10:10 Invited talk - The future of metabolic imaging

Speaker: Dennis Klomp (University Medical Center Utrecht)

MetabolicImagingFuture_Elba_Klomp_May2022.pptx

10:50 → 11:20 Coffee break

11:20 → 12:40 New technologies for PET/MR and TB-PET: Part 1

Convener: Antonio J. Gonzalez

11:20 A highly multiplexed detector readout scheme for Total-Body PET

Nowadays, the technology associated to Time-of-Flight (TOF) Positron Emission Tomography (PET) detectors is reaching the technological limit imposed by the scintillator crystal and photosensors available. Improving the TOF resolution is key to further increase the effective sensitivity of the system. As an alternative, this effective boost in sensitivity can be accomplished by increasing the solid angle coverage of the PET scanner building large axial length systems. This are the so-called Total Body PET (TB-PET) scanners.
TB-PET systems are commercially available, and the first clinical studies have been reported. These systems are the so-called Biograph Vision from Siemens with remarkable TOF capabilities of 217 ps, and the uExplorer from United Imaging. Nevertheless, from the technical perspective, there is still room for further improving these systems by including depth of interaction (DOI) information to provide homogeneous spatial resolution in the entire Field of View (FOV), and/or readout channel reduction since the large number of signals to be handle is a major concern to develop these long axial scanners.
In this work, we show the design and preliminary results of a detector geometry that allows one to mitigate these major concerns on TB-PET systems: the reduction of readout channels as well as the DOI, while providing <300 ps CTR and sub-3mm spatial resolution.

Speaker: David Sanchez (Universidad Politécnica de Valencia)

PSMR_DavidSanchez_V2.pptx

11:40 Performance of monolithic BGO-based detector implementing a Neural-Network event decoding algorithm for TB-PET applications

Total body positron emission tomography (TB-PET) is currently changing the way the medical imaging community approaches PET scanner and data acquisition system designs, shifting the focus to reducing costs while still achieving good 3-D spatial resolution, time resolution and sensitivity.
Monolithic BGO-based detectors allow to significantly cut costs compared to the 3x more expensive standard pixellated LYSO, but it has seldom been used in recent PET scanners because of its slower response and lower light yield, leading to worse time and energy resolution, respectively.
We will present results that disprove this theory, showing that a 25 mm x 25 mm x 12 mm monolithic BGO crystal read by a 4x4 matrix of Hamamatsu S14160 6 mm MPPCs can achieve sub-mm spatial resolution, sub-300 ps CTR and an energy resolution around 15%.
These results are obtained using a novel algorithm for event characterization based neural networks. The algorithm is light and has a real-time implementation in a low-cost FPGA that allows it to process 1 Mcps.
These characteristics make monolithic BGO the optimal scintillator choice for TB-PET, allowing to keep costs down without compromising on performance.

Speaker: Pietro Carra (Istituto Nazionale di Fisica Nucleare)

talk.pptx

12:00 Deep learning for time estimation in monolithic PET detectors using digitized readouts

We perform a simulation study to investigate the potential of convolutional neural networks (CNNs) for gamma arrival time prediction in monolithic PET detectors with waveform digitizers. GATE v8.2 is used to simulate gamma interactions and production of scintillation light in a 50x50x16 mm³ monolithic LYSO crystal, coupled to an 8x8 readout array of silicon photomultipliers (SiPMs). The SiPM waveforms are then simulated as a sum of bi-exponential functions, where we include additional sources of noise such as dark counts. The waveforms are simulated at varying sampling rates from 1 GS/s to 20 GS/s. A 3D CNN is trained to predict the gamma arrival time in the crystal from the leading edge portion of the matrix of detector waveforms, resulting in a coincidence time resolution (CTR) of 140 ps full width at half maximum (FWHM) for the fastest sampling rate. This is a 24% improvement compared to the conventional methodology of leading edge discrimination after baseline correction, followed by an averaging of the first few timestamps, resulting in a CTR of 173 ps FWHM at the same sampling rate. In addition, the 3D CNN maintains respectable timing performance over a large range of sampling rates, only degrading to 191 ps FWHM at 1 GS/s.

Speaker: Jens Maebe

presentation.pptx

12:20 Reducing Memory Requirements for Gradient Tree Boosting Models in FPGA Implementations for Position Estimation in PET Detectors

In PET systems, processing sensor data, such as gamma event positioning, directly on detector level is beneficial for early data reduction and scalability to large systems. Especially in total-body PET with its large amount of detectors, low-cost electronics for early data processing are required. Gradient tree boosting (GTB) has been proven as an accurate method for positioning in planar and DOI direction in radiation detectors. GTB is a supervised machine learning technique based on building ensembles of independent binary decision trees of a pre-defined depth in an additive manner. Input data traverses the trees to the leave nodes, where the path is determined by comparisons of one input data feature to a split value in each tree node. We have shown an FPGA implementation with good positioning performance, high throughputs and low FPGA logic resource consumption. Since memory resources are scarce in low-cost FPGAs, we investigate a method to reduce the memory requirements of GTB models, which can still be high in the FPGA implementation for larger tree depths. GTB models were trained using data acquired with a pixelated high-resolution LYSO scintillator with a 1 mm pitch coupled to a sensor array consisting of 16 digital SiPMs (DPC-3200-22, Philips Digital Photon Counting) with 64 photon channels. From these data, GTB models with varying model hyperparameters were trained for five different feature sets, where the feature sets consist of different combinations of the raw photon counts of the channels and calculated features of the light distribution. For each model, memory requirements were reduced by finding similar split values in the decision tree nodes of the trained models and assigning a common value to these nodes, so that no comparison results are changed. Reducing the amount of input features improves both positioning performance and split value reduction. With no loss of positioning performance, we achieve maximum reductions of split values of more than 50 %. The optimization algorithm is also run with fractions of the training data instead of the full data set, which improves the reduction of split values for a model of 100 trees from 42.5 % to around 60 %, while maintaining more than 99.8 % of positioning performance. With this reduced amount of individual split values, we can reduce the memory requirement of the FPGA implementation of GTB models and employ it on smaller FPGAs, which can reduce the potential cost of total-body PET systems.

Speaker: Karl Krueger (Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, RWTH Aachen University, Aachen, Germany)

2022_PSMR_Krueger_v2.pptx

12:40 → 15:30 Lunch break

15:30 → 16:15 Industrial Session

Convener: Matteo Morrocchi

15:30 Bruker: advantages of industry collaborations

What are the advantages of industry collaborations?
This talk focuses on the advantages that collaborations between academia and industry can bring, and highlights some of the collaborations Bruker BioSpin have done to advance PET imaging in small animals and improve the performance of the scanners.

Moneta_PSMR-Advantages of Industry Collaborations.pptx

15:45 CAEN: tools for discovery meet educational labs!

CAEN S.p.A. is an important industrial spin-off of the INFN (National Institute for Nuclear Physics), since 1979. CAEN brings the experience acquired in almost 40 years of collaboration with the High Energy &
Nuclear Physics community into the University educational laboratories by providing modern physics experiments based on the latest technologies and instrumentation.
CAEN realized different modular Educational Kits, developed together with the University of Insubria (IT) and University of Aveiro (PT), and a series of experiments with different difficulty level. The Educational Kits, provided with a dedicated Control Software, are modern and flexible platforms for teaching the fundamentals of Statistics, Particles Detection and Nuclear Imaging.
The goal is to inspire students and guide them towards the analysis and comprehension of different physics
phenomena with a series of experiments based on state-of-the-art technologies, instruments and methods.
An example of CAEN kit is the EasyPET, a user friendly and portable PET system, designed to explore the physical and technological principles of the conventional human PET scanners. The device uses only two detectors to execute a PET scan, simplifying the set-up to make it accessible to Educational Laboratories.
A Graphic User Interface allows the user to set the acquisition parameters, visualize the reconstructed image in real time during acquisition and perform several didactic experiments related to PET imaging, as well as offline image analysis

CAEN_PSMR-TBP 2022 Workshop.pptx

Catalog

16:00 Positrigo: a dedicated brain PET system and its applications

Positrigo has developed the cost-effective dedicated brain PET System NeuroLF. We will present preliminary design properties and potential applications in research and clinical routine.

16:15 → 17:30 Poster session

Conveners: Alberto Del Guerra, Juan José Vaquero

16:15 Screen 01 - Investigation of the Impact of MR Hardware Attenuation on TOF and non-TOF PET/MR Images

The combination of positron emission tomography (PET) and magnetic resonance imaging (MR) allows the simultaneous scanning of a patient's anatomy as well as physiology. This setup comes with difficulties, as the coil hardware (HW) that is used for MR-acquisition attenuates the photons which are acquired by PET. Disregarding this attenuation can lead to an underestimation of accumulated activity in the patient.
This study investigates the influence of different attenuating hardware and phantoms using simulation. An activity cylinder was attenuated by headphones (either foam or PVC) or a multichannel MR-coil using the CT image of the latter. Phantom attenuation was changed between water and air to investigate its influence in iterative reconstruction methods. Additionally, the attenuating effect of PVC-headphones on brainweb data was investigated. As earlier studies have shown that TOF emission data contains substantial information about the attenuation, simulations were performed utilising both TOF and non-TOF data.

Speaker: Nicole Jurjew

picture_Jurjew.jpg

poster_Jurjew.pdf

summary_Jurjew.pdf

16:15 Screen 02 - Assessing the robustness of radiomics feature measurements using the noise equivalent count rate, and the future role of Total Body PET

Radiomics, the term given to a feature-based approach to image analysis, suffers when used with inherently noisy PET data due to the convoluted derivation of many features and an overall lack of uncertainty provision. The noise equivalent count rate (NECR) is a scanner- and geometry-dependent metric which is primarily used to verify scanner performance. The characteristic variation of the NECR with activity in the field of view prompts discussion over its potential use to predict radiomics feature values at injected activities with an optimum signal-noise ratio. Our study began by scanning a cylindrical phantom filled with FDG solution for alternating 5 and 25 minute acquisitions, presenting 48 total frames. This was repeated twice using our own custom 3D-printed anthropomorphic tumour phantom inserts for the NEMA Image Quality phantom. 32 radiomics features from texture matrices were observed to correlate well (Pearson product moment correlation coefficient ($|\textrm{PMCC}| \geq 0.9$) with NECR for the 25 minute frames, enabling estimation of 'compensation factors' from the values at clinical injection activity levels. Correlations with NECR fall across all metrics for 5 minute scans, with the $|\textrm{PMCC}|$ of the ten previously highest-correlating features falling by an average of $(11.5 \pm 6.6)~\%$. Feature correlations to 'global' NECR also decrease for small 'local' ROIs as observed in our custom phantom scans. Total Body PET systems have reported a 4-fold increase in sensitivity in comparison with conventional scanners for activity in the field of view, such that equivalent scan signal-noise ratio can be achieved in one quarter of the total scanning time. This study considers how the dual influence of increased sensitivity and choice of noise-representative metric may lead to a method for estimating uncertainty on these radiomics features.

Speaker: George Needham

picture_gneedham_psmrtbp22.jpg.jpeg

poster_GNeedham_PSMRTBP2022.pdf

summary_gneedham_psmrtbp22.pdf

16:15 Screen 03 - Synergistic Image Reconstruction Framework: version 3.2

The Synergistic Image Reconstruction Framework (SIRF) is a research tool for reconstructing data from multiple imaging modalities, currently most prominently PET and MR. Included are acquisition models, reconstruction algorithms, registration tools, and regularisation models. In this work, we briefly list current capabilities and demonstrate the main new feature added since SIRF 3.1: acquisition models for non-Cartesian MR sequences.

The new reconstruction capabilities were tested on three 2D MR datasets that were acquired on a Siemens 3T scanner using the open-source MR pulse design framework pulseq. Data were acquired with a cartesian, golden-angle radial and spiral trajectory and reconstructed using the acquisition models SIRF.

The presented work shows that SIRF was able to reconstruct the acquired data using both a pseudo-inverse as well as an iterative reconstruction algorithm independent of the employed sampling pattern.

The new functionality makes SIRF more flexible with respect to MR input data as well extends its potential user base.

Speaker: Kris Thielemans (Institute of Nuclear Medicine, University College London, London, UK)

CCP SyneRBI SIRF download page

picture_kthielemans.jpg

poster_SIRFv32.pdf

summary_SIRFv32.pdf

16:15 Screen 04 - A Convolutional Neural Network for Automated Delineation and Classification of Metabolic Tumor Volume in Head and Neck Cancer

Deep Learning based approaches for automated analysis of tomographic image data are drawing ever increasing attention in Radiology and Nuclear Medicine. With the advent of the new generation of PET scanners with massively enlarged axial field of view (“total body PET”) the importance of integrating such approaches into clinical workflows will further increase. In the present study we report on our application of a convolutional neural network (CNN) for automated survival analysis in head and neck cancer (HNC): PET parameters such as metabolic tumor volume (MTV), total lesion glycolysis, and asphericity of the primary tumor are known to be prognostic of clinical outcome in HNC patients. Additionally including evaluation of lymph node metastases further increases the prognostic value of PET. However, accurate manual delineation and classification of all lesions is time consuming and incompatible with clinical routine. Our goal, therefore, was development and evaluation of an automated tool for MTV delineation/classification of primary tumor and lymph node metastases in HNC in PET.
Automated delineation of the HNC cancer lesions was per- formed with a residual 3D U-Net convolutional neural network (CNN). 698 FDG PET/CT scans from 3 different sites and 4 public databases were used for network training and testing. In these data, primary tumor and metastases were manually delineated (with assistance of semi-automatic tools) and accordingly labeled by an experienced physician. Performance of the trained network models was assessed by 5-fold cross validation using the Dice similarity coefficient for individual delineation tasks.
Additionally, survival analysis using univariate Cox regression was performed. Delineation of all malignant lesions with the trained U-Net model achieves a Dice coefficient of 0.866 when not discriminating between primary tumor and lymph nodes. Treating primary tumor and lymph node metastases as distinct classes yields Dice coefficients of 0.835 and 0.757 for the respective delineations. The univariate Cox analysis reveals that, both, manually as well as automatically derived total MTVs are highly prognostic with similar hazard ratios (HR) with respect to overall survival (HR=1.8; P<0.001 and HR=1.7; P<0.001, respectively). To the best of our knowledge, our work represents the first CNN model for successful MTV delineation and lesion classification in HNC. The network quickly performs usually satisfactory delineation and classification of primary tumor and lymph node metastases in HNC using FDG-PET data alone with only minimal sporadic manual corrections required. It is able to massively facilitate study data evaluation
in large patient groups and also does have clear potential for clinical application.

Speaker: Dr Pavel Nikulin (Helmholtz-Zentrum Dresden-Rossendorf, PET Center, Institute of Radiopharmaceutical Cancer Research)

picture_nikulin.png

poster_nikulin.pdf

summary_nikulin.pptx

16:15 Screen 05 - iPET - Current developments and improvements of preclinical PET scanners based in easyPET technology

The iPET prototype, based on the EasyPET technology, is an affordable preclinical PET scanner capable of real-time high quality in-vivo images. While the high price of preclinical PET scanners makes them unaffordable to many research centers, easyPET-based systems, using an innovative scanning method based on two rotation shafts for the movement of detector arrays, reduce the overall costs without compromising image quality.
High and uniform spatial resolution over the whole field of view (FoV) is achieved, by minimizing scattered events and parallax errors due to depth of interaction (DoI) uncertainty. Setting the detector modules very close to the subject, favouring sensitivity, is possible by software. Full body mouse imaging is possible using only a small number of detector elements, capable of scanning billions of lines of response (LoRs) in few minutes.
The prototype uses two arrays of 300 detector cells (1.5 x 1.5 x 20 mm3 LYSO scintillators coupled to SiPMs), covering up to 100 mm diameter × 80 mm long FoV. A dedicated frontend board processes the coincidence signals based on a fast dual channel ADC (200 MHz) and FPGA for data acquisition. The system can digitize the full pulse waveform and perform the pulse height measurement and coincidence sorter on computer. Multiplexing Anger logic is used for readout to simplify electronics and 511 keV events flood map shown for a group of 25 detectors.
LoRs are organized in a List-Mode format and processed by a dedicated algorithm based in Maximum-Likelihood Estimation Methods (LM-MLEM), running in GPU CUDA kernels, delivering results in few seconds – 10^11 updated voxels/second. This speed gain allows real-time visualization of the reconstructed images during in- vivo PET scanning, with multiple advantages such as the possibility of identification and correction of mispositioning of the animal in the scanner FoV.
GATE simulations using MOBY phantom are shown, where activity was distributed in four regions (kidneys, heart and thyroid) and with background activity, demonstrating the preclinical capabilities using a few number of detectors.

Speaker: Mr Pedro Manuel Mendes Correia (University of Aveiro)

photo_PC.jpg

poster_iPET-PC.pdf

summary_iPET-PC.pdf

16:15 Screen 06 - SAFIR-I: First Time-Activity Curves in Rat Brain in vivo

The brain of an in vivo rat has been imaged with 18F-Fluorodeoxyglucose (FDG) by means of the SAFIR-I PET insert, at injected activities of 29.6MBq and 314.6MBq. A measurement of a 50ml Falcon tube filled with a solution of FDG and water at 50MBq was used as approximate quantitative calibration of image voxel values. Based on the data, time-activity curves (TACs) have been generated for the cortex region and the whole brain at low and high decay rate. The curves are consistent with biological expectations. SAFIR-I could demonstrate its ability to acquire meaningful quantitative PET images at injected activities stretching beyond 300MBq. The accuracy of the TACs is still expected to improve with the inclusion of attenuation, random, normalization and scatter corrections, as well as quantitative calibration with a dedicated calibration phantom measurement at a comparable decay rate.

Speaker: Mr Pascal Bebié (ETH Zurich, Institute for Particle Physics and Astrophysics)

Picture_SAFIR-I_TAC_PSMR22.jpg

Poster_SAFIR-I_TAC_PSMR22_v1.pdf

summary_SAFIR-I_TAC_PSMR22.pptx

16:15 Screen 07 - Monte Carlo evaluation of a total-body rat preclinical scanner based on AI-enhanced BGO detectors.

This work presents the Monte Carlo characterization of a total-body rat preclinical scanner obtained through experimental detectors characterization and simulated phantom
acquisitions. The proposed scanner scanner has a 112.8 mm inner diameter and 357 mm axial FOV which is able to cover the whole rat body. The scanner architecture consists of 7 rings each of which has 8 BGO detectors. Each detector consists of a 51×51×12 mm3 BGO crystal read by 6 mm SiPMs by Hamamatsu. Recent works have shown that BGO performance can be enhanced with the use of AI algorithms. In the simulation pipeline, several parameters obtained through experimental detectors characterization have been included. Among these are the detector spatial resolution of 1.1 FWHM (x,y) and 1.8 mm FWHM(DOI), an energy resolution of 20.2 % and a temporal resolution of 300 ps. NEMA NU4-2008 image quality phantoms have been simulated. The image quality results are reported in terms of noise, recovery coefficient (RC) and spill-over ratio (SOR) while the performance in terms of spatial resolution have been investigated using a Derenzo phantom. The histogram-mode data were reconstructed by using ordered-subset-expectation-maximization (OSEM) algorithm.

Speaker: Dr Luigi Masturzo (Istituto Nazionale di Fisica Nucleare)

picture_summary_poster

• picture_LuigiMasturzo.tiff
• poster_LuigiMasturzo.pdf
• summary_LuigiMasturzo.pptx

16:15 Screen 08 - ScintoTube: An Edgeless preclinical PET insert

Positron Emission Tomography (PET) research in the small animal (pre-clinical) field is driven by improving spatial resolution and sensitivity. Conventional PET scanners are built of multiple detector modules placed in a cylindrical geometry. The unavoidable gaps between the detector modules decrease sensitivity and degrade spatial resolution towards the edges of the system field of view (FOV). To mitigate the modular design associated problems, it has been proposed to design edge-less scanners. Following this idea, we already designed and validated an edgeless pre-clinical PET insert based on a single LYSO annulus with a cylindrical inner diameter but 10 outer facets. The system provided good performance but some undesirable effects were observed in the light distribution (LD) pattern in the annulus joints. To account for this, we have modified the design and already built a new version of this edgeless scanner in which both outer and inner facer are cylindrical. We present here the preliminary evaluation of the system performance to demonstrated that we have solved the problems encountered in the first design.

Speaker: Marta Freire (Institute for Instrumentation in Molecular Imaging (i3m))

Picture_PSMR2022MartaFreire.png

Poster_PSMR2022Tube_FreireVertical.pdf

Summary_PSMR2022_Tube.pptx

16:15 Screen 09 - Simulation Results for the PETITION PET Systems

The PETITION (PET for InTensive care units and Innovative protON therapy) collaboration is currently developing two Positron Emission Tomography (PET) systems with a focus on new clinical applications. Both PET scanners are dedicated brain systems with 179 mm axial field of view and an inner diameter of 300 mm. The first system (ICU-system) is a mobile PET scanner. It will be moved to the patients in the intensive care units (ICU), who cannot be transferred to conventional imaging facilities. The second system (PBT-system) is designed for online PET imaging in proton beam therapy (PBT). Its main difference to conventional PET scanners is an opening to allow for proton irradiation during imaging.

We developed a modular hardware system with identical readout modules for both PET systems. These comprise blocks of Lutetium-yttrium oxyorthosilicate scintillators, each with 5 × 5 crystals of 2.74 × 2.74 × 15 mm$^3$. For this study we perform a series of Monte-Carlo-Simulations with the Gate software package, to obtain estimates for the sensitivity profiles and spatial resolution.

Before coincidence search (2 ns window) the singles are filtered with an energy window (390 keV − 610 keV). We measure peak sensitivities of 3.19 % (ICU-system) and 1.54 % (PBT-system). The spatial resolution varies between 1.8 mm and 2.7 mm for the central region of both PET-systems. This is in agreement with the theoretical expectation for such scanners.

Speaker: Dr Christian Ritzer (Institute for Particle Physics and Astrophysics, ETH Zürich, Switzerland)

picture_Ritzer.JPG

poster_Ritzer.pdf

summary_Ritzer.pdf

16:15 Screen 10 - Simulation Study of an On-Chip PET System

Organs-on-Chips (OOCs), microdevices mimicking in vivo organs, find growing applications in disease modeling and drug discovery. With the increasing number of uses comes a strong demand for imaging capabilities of OOCs. Positron Emission Tomography (PET) would be ideal for OOC imaging, however, current PET systems have insufficient spatial resolution for this task.
In this work, we propose the concept of an On-Chip PET system capable of imaging OOCs. Our system consists of four detectors arranged around the OOC device. Each detector is made of two monolithic Lutetium–yttrium oxyorthosilicate (LYSO) crystals and covered with Silicon photomultipliers (SiPMs) on multiple surfaces. We use a Convolutional Neural Network (CNN) trained with data from a Monte Carlo Simulation (MCS) to predict the first gamma-ray interaction position inside the detector from the light patterns that are recorded by the SiPMs on the detector’s surfaces. With the Line of Responses (LORs) created by the predicted interaction positions, we reconstruct with Simultaneous Algebraic Reconstruction Technique (SART).
The CNN achieves a mean average prediction error of 0.78 mm in the best configuration. We use the trained network to reconstruct an image of a grid of 21 point sources spread across the field of view and obtain a mean spatial resolution of 0.53 mm.
We demonstrate that it is possible to achieve a spatial resolution of almost 0.5 mm in a PET system made of multiple monolithic LYSO crystals by directly predicting the scintillation position from light patterns created with SiPMs.

Speaker: Christoph Clement (Inselspital Bern)

picture_clement.jpg

poster_clement.pdf

summary_clement.pdf

16:15 Screen 11 - The Time of Flight PET for Proton Therapy (TPPT)

The recent rise of Proton Therapy as a cancer treatment option motivates the need for sophisticated imaging to monitor the position of the proton beam deposition in a patient. Using Time-of-Flight (ToF) PET imaging, the project seeks to provide information of beam location and intensity in a patient by monitoring the presence of positron emitting isotopes produced due to energy deposition of the proton beam, known commonly as Proton-Range Verification.

Speaker: Kyle Klein (University of Texas at Austin)

Picture_Klein.jpg

Poster_Klein.pdf

Summary_Klein.pdf

16:15 Screen 12 - Preliminary Monte Carlo study of Modular J-PET

Modular J-PET as the latest prototype of J-PET collaboration, utilizes the unique detection principles thanks to its distinguished detector arrangements. It consists of 24 detection units called modules which are arranged in regular 24-sided polygon circumscribing a circle which provides 50 cm of AFOV. The one advantages of this prototype is an arrangement of scintillators and photo-detectors which provides possibility of performing portable imaging that improves functionality of tomograph. The presented study was carried out by Geant4 application for tomographic emission (GATE)
simulation toolkit. The sensitivity and scatter fraction as two main characteristics of tomographs will be investigated according to the National Electrical Manufacturers Association norm(NEMA NU2-2018).
In this study, the scatter fraction and sensitivity of Modular J-PET have been investigated according to the NEMA NU2-2018 standards. The study was performed by the GATE simulation package. The performed simulations indicate that the peak of sensitivity for Modular J-PET was 4 cps/kBq and the sensitivity was around 1.5 cps/kBq. The Scatter Fraction was calculated based on SSRB algorithmes with the amount of 3.64%. Although the AFOV of Modular j-PET is larger than the standards PET, obtained values of scatter fraction is similar to those computed for commercial PET scanners. For example, the GE Discovery has a scatter fraction of between 21% and 34%, dependent on the mode used and sensitivity for the GE Discovery is 20.4 cps /kBq.

Speaker: Faranak Tayefi Ardebili (1- Marian Smoluchowski Institute of Physics, Jagiellonian University, Poland. 2- Total-Body Jagiellonian-PET Laboratory, Jagiellonian University,Poland)

Picture.jpg

Poster.pdf

Summary.pdf

16:15 Screen 13 - TOF MLEM for Total-Body J-PET with Analytical System Response and Resolution Modelling

We report the reconstructed results of the simulated NEMA IEC and static XCAT phantoms, obtained using the original time-of-flight maximum likelihood expectation maximisation (TOF MLEM) algorithm, designed for total-body modular Jagiellonian PET (J-PET) scanners. The continuous J-PET detectors allow the system response matrix (SRM) to be defined as a set of log-polynomial models, derived from fitting the Monte Carlo simulated emissions of annihilation photons on 2D planes with differing obliqueness. The new TOF MLEM accounts for the resolution modelling in detector space, parallax effect and smearing in J-PET scintillator strips. Compared to the algorithm with no resolution model, a considerable improvement in image quality and other metrics was observed. We also explore a simplified attenuation correction using on-the-fly weighting for each measurement, which proved to be less sensitive to boundary effects and produced similar or better results than the traditional integration over bins. The proposed analytical SRM model for the total-body J-PET can be further upgraded to account for the non-collinearity, positron range and other factors.

Speaker: Roman Shopa (National Centre for Nuclear Research, Poland)

picture_Shopa.png

poster_Shopa.pdf

summary_Shopa.pdf

16:15 Screen 14 - Influence of Spatial Resolution and SNR of Attenuation Correction Maps on Breast PET images in a Fully-Hybrid PET/MR system

This preliminary work investigates how PET images are affected both quantitatively and qualitatively by the modification of spatial resolution and SNR of the MR sequences for AC (MRAC). Forty-six women with breast cancer underwent 18F-FDG PET/MR study using the 3T SIGNA PET/MR (General Electric Healthcare, Waukesha, WI, US) scanner. During breast prone PET scan, standard MRAC acquisition has been collected. Additionally, five different MRAC sequences were acquired with the same parameters of the standard MRAC (reference) except for parameters related to spatial resolution, SNR and shimming. Using offline GE Duetto Toolbox, six PET reconstructed images were obtained correcting attenuation with different MRACs. Eighty-three lesions (46 breast lesions and 37 active lymph nodes) were identified on the reference PET reconstructed images. SUVmax and SUVmean were calculated together with the percentage differences (SUVdiff) and the Root Mean Squared Error (RMSE) between the new set of images and the reference. Results from our work show that spatial resolution and SNR acquisition parameter modifications in MRAC sequences seem to affect SUVmean and SUVmax of the corresponding corrected PET images, but not significantly. The MRAC sequence acquired with improved shimming on the dorsal side has shown the highest effect on the RMSE both for AC-maps and for PET images.

Speaker: Dr Paola Scifo (Nuclear Medicine Dept, IRCCS San Raffaele Scientific Institute, Milan, Italy)

picture-scifo.jpeg

poster-scifo-breast.pdf

summary-scifo-breast.pptx

16:15 Screen 15 - Attenuation Correction in PET/MRI pediatric Brain Tumors: a preliminary comparison between the ZTE and Atlas techniques

The use of simultaneous hybrid PET/MRI is highly beneficial in children as it allows for a one-stop examination that limits the scanning procedures, the need for a second sedation and also reduces the radiation dose compared to PET/CT. The attenuation correction, in the brain, is performed with Atlas, with Zero Echo Time/Ultrashort Echo Time (ZTE/UTE) sequences or, more recently, with Deep Learning-based methods. In pediatric patients, and in particular in oncological pediatric studies where tumor resection may occur, the size and the abnormal shape of the head maybe critical for a good attenuation correction. In this work, we have compared the two Atlas and ZTE attenuation correction methodologies in a group of treated pediatric patients with brain tumors who underwent 11C-Methionine PET/MRI scans. This preliminary work shows that the use of ZTE or Atlas for the generation of ACmaps can modify the SUVs of the study but the differences in the values are very small. The DICE indexes calculated between lesions VOIs are high. Finally, it should be noticed that some of the patients of our sample have external ventricular drain catheter valves that generate artifacts. In these cases, interestingly, both ZTE and Atlas show bulk of signals but ACmapAtlas has smaller artifacts than ACmapsZTE.

Speaker: Dr Paola Scifo (Nuclear Medicine Dept, IRCCS San Raffaele Scientific Institute, Milan, Italy)

picture-scifo.jpeg

poster-scifo-brain.pdf

summary-scifo-brain.pptx

16:15 Screen 16 - Exploring the Utility of Radiomic Feature Extraction to Improve the Diagnostic Accuracy of Cardiac Sarcoidosis Using FDG PET

This study aimed to explore the radiomic features from PET images to detect active cardiac sarcoidosis (CS). Methods: Forty sarcoid patients and twenty-nine controls were scanned using FDG PET-CMR. Five feature classes were compared between the groups. From the PET images alone, two different segmentations were drawn. For segmentation A, a region of interest (ROI) was manually delineated for the patients’ myocardium hot regions with standardized uptake value (SUV) higher than 2.5 and the controls’ normal myocardium region. A second ROI was drawn in the entire left ventricular myocardium for both study groups, segmentation B. The conventional metrics and radiomic features were then extracted for each ROI. Mann-Whitney U test and a logistic regression classifier were used to compare the individual features of the study groups. Results: For segmentation A, the SUVmin had the highest area under the curve (AUC) and greatest accuracy among the conventional metrics. However, for both segmentations, the AUC and accuracy of the TBRmax were relatively high, greater than 0.85. Twenty-two (from segmentation A) and thirty-five (from segmentation B) of 75 radiomic features fulfilled the criteria: P-value less than 0.00061 (after Bonferroni correction), AUC greater than 0.5, and accuracy greater than 0.7. Principal Component Analysis (PCA) was conducted, with five components leading to cumulative variance higher than 90%. Ten machine learning classifiers were then tested and trained. Most of them had AUCs and accuracies ≥ 0.8. For segmentation A, the AUCs and accuracies of all classifiers are greater than 0.9, but k-neighbors and neural network classifiers were the highest (=1). For segmentation B, there are four classifiers with AUCs and accuracies ≥ 0.8. However, the gaussian process classifier indicated the highest AUC and accuracy (0.9 and 0.8, respectively). Conclusion: Radiomic analysis of the specific PET data was not proven to be necessary for the detection of CS. However, building an automated procedure will help to accelerate the analysis and potentially lead to more reproducible findings across different scanners and imaging centers and consequently improve standardization procedures that are important for clinical trials and development of more robust diagnostic protocols.

Speaker: Prof. Charalampos Tsoumpas

Poster_NM.pdf

Summary_NM.pptx

17:30 → 18:00 Coffee break

18:00 → 19:00 New technologies for PET/MR and TB-PET: Part 2

Convener: Giancarlo Sportelli

18:00 PETAT1, a Time-Sorting Readout ASIC for PET

Modern PET scanners are often based on scintillating crystals read out by photo sensors. The large number of channels is processed by specialized electronic microchips. Data produced by these chips must be received by suitable data acquisition circuits and brought outside of the scanner to a data processing unit where hits must be sorted in time in order to find pairs of hits within a coincidence time window. The components required to transport the data (often FPGAs) must be located close to the sensors in large systems where they consume space and require power and cooling. We propose a novel readout architecture where no auxiliary readout circuits are required and which in addition outputs all hits already sorted by their time-stamps. This is achieved by adding data inputs to each readout chip so that events can be sent from chip to chip. The incoming external hits and the internal hits are buffered and merged by taking into account the hit time information so that the oldest hit is sent out first. The outgoing time-sorted hit data stream can then be merged efficiently with further streams in downstream chips. The architecture has been simulated in full detail with PET-like data. Hit losses are small up to the bandwidth limit of the serial link. The prototype chip PETAT1 including a SiPM readout with amplitude and time measurement has been designed, produced and operated.

Speaker: Prof. Peter Fischer (Heidelberg University)

2022_PSMR_Fischer.pdf

2022_PSMR_Fischer.pptx

18:20 Sub-100 ps Coincidente Time Resolution for ToF-PET detectors using FastIC

Recent developments in Silicon Photomultiplier (SiPM) sensors and fast readout electronics are enabling new achievements in Coincidence-Time Resolution (CTR) for Positron-Emission Tomography (PET) detectors. The new 8-channel FastIC ASIC developed in CMOS 65~nm technology provides an accurate time-stamping of the detected particles and a linear energy measurement while having a power consumption of 12~mW/Ch. The ASIC have the capability of summing different readout channels in clusters of 4 channels before measuring time and energy. Evaluation of the ASIC showed CTR values of 94 ± 2 and 76 ± 2 ps FWHM for HPK SiPM S13360-3050PE and new technology FBK HD-NUV Low Field (3.2 x 3.12 mm² pixel, 40 μm cell) coupled to a pair of LSO:Ce:Ca 0.2% measuring 2 x 2 x 3 mm³. Furthermore, new FBK HD-NUV Low Field SiPM coupled to LYSO:Ce:Ca 0.2% measuring 3.13 x 3.13 x 20 mm³ yielded a CTR value of 126 ± 2 ps FWHM. Additionally, first measurements of the chip confirmed its ability to detect prompt light emitters e.g., using Cherenkov radiators like TlCl and PbF₂.

Speaker: Antonio Mariscal-Castilla (Departament de Física Quàntica i Astrofísica, Institut de Ciències del Cosmos, University of Barcelona, Barcelona, Spain)

20220512_PSMRTB_ANTONIOMARISCAL-CASTILLA.pdf

18:40 Blumino: a fully integrated analog SiPM with on-chip time conversion

Blumino is a fully integrated analog silicon photomultiplier with on-chip discriminator and time-to-digital converter. The benefit of this approach is the scalability and compactness. In addition, the small form factor, high gain and low parasitic capacitance improve the overall timing performance. The sensor comprises an analog SiPM developed by On Semiconductor which exhibits a PDE greater than 40 % at 420 nm and DCR of 81.7 kcps/mm2. The total sensitive area is 3 mm x 3 mm with a fill factor of 75 %. The integration process combined custom SPAD processes with standard CMOS in a 350 nm technology node. A compact board was developed to take advantage of the sensor characteristics (separate timing and energy channels) and to allow tiling for a more complex coincidence system. The sensor board comprises a transimpedance amplifier, a fast comparator, a charge integrator and adjustable supply and bias voltages, all controlled by a microcontroller that also offers pre-processing capabilities. The current system offers great potential for different coincidence configurations of various dimensions.

Speaker: Mrs Andrada Muntean (EPFL)

AndradaMuntean_PSMRTBP_FINAL.pptx

20:00 → 22:00 Conference dinner

Tuesday, 31 May

08:30 → 10:30 New technologies for PET/MR and TB-PET: Part 3

Convener: Paul Lecoq

08:30 RF shielding on scintillator level for highly-integrated PET/MRI systems

The combination of PET and MRI to a simultaneous system poses challenges concerning the integration and optimization of both subsystems' performance. One crucial part of the integration process is a good RF shielding of the PET modules to reduce mutual interferences. The PET modules are typically completely enclosed and restrict the system configuration. We propose an approach with RF shielding on scintillator level to achieve a higher level of integration and to reduce the PET module's RF housing size. The scintillator is excluded from the shielded volume as it is not influenced by the electromagnetic fields, and the created opening is closed using the scintillator with integrated RF shielding material to maintain the shielding effectiveness (SE). This is done by wrapping copper foil around the scintillator segments to create waveguides operating below their cut-off frequency. To prove the feasibility, prototypes were assembled using 2 $\times$ 32 mm$^2$ PVC slabs with a height of 7 or 12 mm, and 12.5 $\mu$m copper foil inserted between every or every second slab. The directional SE was evaluated using a network analyzer and field probes, where the receive field probe surface normal was oriented in segmented, monolithic and DOI direction, analog to the orientations used for the PET detector performance evaluation. The influence of the integrated material on the PET detector performance was evaluated using semi-monolithic LYSO arrays (eight 3.9 $\times$ 32 $\times$ 12 mm$^3$ slabs) and digital SiPM arrays by comparing the performance of three scintillator configurations: Slab$\mathrm{_{ESR}}$ with a double layer of ESR as crystal separation; Slab$\mathrm{_{ESR+Cu}}$ with a double layer of ESR and copper foil inserted in-between; and Slab$\mathrm{_{Cu}}$ with only copper foil wrappings. The positioning performance was evaluated using a fan-beam-collimator setup and the machine learning technique gradient tree boosting. The energy and timing resolution were evaluated using flood irradiation measurements without a collimator.

The prototypes with copper foil between each slab reached the highest attenuation for 12 mm height (SE of about 20 dB, 0.3 dB and 6 dB, in segmented, monolithic and DOI orientation). We achieved the overall best positioning performance for Slab$\mathrm{_{Cu}}$, followed by Slab$\mathrm{_{ESR}}$ and Slab$\mathrm{_{ESR+Cu}}$. The best energy resolution was seen for Slab$\mathrm{_{ESR}}$, followed by Slab$\mathrm{_{ESR+Cu}}$ and Slab$\mathrm{_{Cu}}$ (10.6%, 11.2% and 12.6%, respectively). For the timing resolution, Slab$\mathrm{_{ESR}}$ (279 ps) was followed by Slab$\mathrm{_{Cu}}$ (288 ps). Slab$\mathrm{_{ESR+Cu}}$ performed worst with 293 ps.

In summary, we have seen a minor impact of the introduced copper foil on the PET detector performance. In combination with the achieved SE, we have shown the feasibility of this shielding approach and have found two potential integration methods for system applications (integration between a double layer of ESR foil or directly wrapping scintillator segments with copper foil).

Speaker: Laiyin Yin (Department of Physics of Molecular Imaging Systems, Institute for Experimental Molecular Imaging, RWTH Aachen University)

20220530_PSMR_Talk_YinGrossmann.pptx

08:50 Geometrical considerations on hexagonal SiPM

The ideal distribution of the detectors of a preclinical Positron Emission Tomography scanner is a sphere with the subject located at its center. However, rigid scintillators crystals and detectors do not suit smooth figures. A truncated icosahedron has been previously proposed as a feasible approach to cover 4$\pi$ steradian. In order to tile the polyhedron, hexagonal scintillators and photodetectors have been designed and manufactured. In this work, we explore the advantages of the hexagonal geometry. Four scintillator geometries (hexagon, square, triangle and cylinder) are compared in terms of their light yield. A field flood image of a hexagonal scintillation array of 181 crystals read with four channels multiplexing along the diagonals of the hexagonal photodetector is presented. Finally, the pixel resolution of a sub-surface laser engraving square scintillator array is evaluated with hexagonal and a square photodetectors.

Speaker: Dr Rigoberto Chil (Universidad Carlos III de Madrid)

Geometrical considerations on hexagonal SiPMs.pdf

09:10 Development and validation of a measurement-driven inter crystal scatter recovery algorithm with in-system calibration

In PET a high percentage of gamma photons being detected undergo Compton scatter in the scintillator. Depending on the angle of incidence and the scintillator geometry this might lead to inter crystal scatter (ICS) events, where energy is deposited in two or more different crystals in the detector, which common positioning and reconstruction algorithms cannot resolve. Therefore, scattered events worsen the spatial resolution and the signal-to-noise ratio in the reconstruction. We want to address this challenge by recovering crystal interactions of an event and feeding them into a reconstruction framework. In this work, we established an algorithm based on a one-to-one coupled detector, which combines a measurement-driven calibration and a fitting routine to achieve the recovery of crystal interactions. Using Geant4 simulations, we validated this approach by showing that crystals and their energies could be recovered to a satisfactory degree.

Speaker: Ms Katrin Herweg (RWTH Aachen University)

Presentation_PSMR_KH_final.pptx

09:30 Realistic Total-Body J-PET geometry optimization - Monte Carlo study

Total-body PET imaging is one of the most exploited topics in medical imaging. State-of-the-art PET scanners use inorganic scintillators such as L(Y)SO or BGO. We report the performance comparative studies of the total-body PET scanners using Jagiellonian PET (J-PET) technology that is based on the plastic scintillators. Four realistic total-body scanner geometries, varied in the number of rings, scanner radius, and distance between the neighbouring rings were considered. Monte Carlo simulations of two NEMA phantoms (2-meter sensitivity line source and image quality) were generated to assess the performance of the tested geometries. Significant differences in the sensitivity of the scanners were observed. At the same time, relatively small differences in the image quality metrics of the reconstructed NEMA IEC phantoms images were found. The optimal scanner design was recommended for the next generation of the J-PET total-body devices.

Speaker: Jakub Baran (Jagiellonian University)

BaranJakubPSMR2022-1.pdf

09:50 Investigation of the DOI capable configuration in dealing with the parallax error in the Total-Body J-PET tomograph

There is an ongoing interest in the development of Total-Body PET scanners. J-PET Collaboration from Jagiellonian University, as one of the groups focused on development of such tomograph, investigate the possibility of PET scanner construction based on the novel geometrical configuration and unique utilization of organic scintillators. Despite many advantages of Total-Body PETs, they create new challenges. The impact of parallax error decreases the system resolution and as a result, the image quality. Solution for that may be introduced by creation of a PET system with depth of interaction (DOI) capability, which can provide more accurate information about the location of gamma photon interaction with the scintillator material. In this simulation-based work a new DOI capable Total-Body PET tomograph designed with the J-PET technology has been inspected. Its feasibility was assessed by comparing to the standard Total-Body J-PET tomograph as a dependence of characteristics such as sensitivity, scatter fraction and coincidences share. Both scanners were simulated using GATE simulation software. Additionally, impact of the 45° acceptance angle has been assessed. As expected, the DOI capable solution does not have influence on investigated characteristics, which suggest its applicability for the parallax error correction. As a next step, the effect of this human-grade Total-Body J-PET configuration on the imaging will be investigated.

Speaker: Meysam Dadgar (Jagiellonian University)

Meysam_Dadgar.pdf

10:10 Hyperion III – MRI-Compatible PET Detector Platform

Commercially available PET/MRI scanner have been designed as whole-body systems. In these, PET spatial resolution and sensitivity are limited. Dedicated PET inserts can potentially overcome these limitations, but every application, e.g., neuro, breast, or preclinical, has different requirements.
A PET detector platform for simultaneous PET/MRI was thus designed, providing the needed flexibility to construct different systems. Different detector technologies shall be usable to detect scintillation light from the PET detector crystals. The first implementation of a sensor tile uses DPC-3200 sensors from PDPC. It offers 144 channels in a sensitive area of approx. 48×48 mm2. The integration of analog SiPM/ASIC combinations, e.g., using the PETSys TOFPET2 ASIC is currently planned. The sensor tiles are connected with flexible cables to the detector mother boards. Cable lengths of up to 5 m were tested successfully. Each mother board supplies up to 15 sensor tiles and transmits their data via 10-Gigabit Ethernet to a data acquisition and processing server. There, the raw data is either stored for later analysis – e.g., for system calibration – or directly processed and saved as a listmode file.
The different PET systems are configured and controlled with a separate control software, providing a graphical user interface.

Speaker: Dr Bjoern Weissler (RWTH Aachen University, Hyperion Hybrid Imaging Systems)

2022-05-27 - PSMR2022 - Hyperion III Platform _ compressed.pptx

10:30 → 11:00 Coffee break

11:00 → 12:20 New technologies for PET/MR and TB-PET: Part 4

Convener: Paweł Moskal

11:00 Perspective for a Total Body PET with ≤100ps timing resolution

The quest for pushing the limits of PET’s effective sensitivity has led to two important R&D lines, illustrated by the remarkable development of the Total Body PET Explorer project and the 10 ps Time-of-Flight PET challenge.
But there is no reason why these two approaches cannot be combined, along a route already open by the Siemens Biograph Vision Quadra machine, with a length of 106cm and a quoted CTR resolution of 228ps at the peak of the NEC sensitivity.
This talk will show how recent progress in the development of metascintillators, combining the high stopping power of well know scintillators with the sub ns emission of cross-luminescent scintillators, or the ultrafast photon emission resulting from the 1D, 2D, or 3D quantum confinement of the excitons in nanocrystals, open interesting perspectives for the development of a Total Body PET with ≤100ps timing resolution.
A trade-off between the axial length and the timing resolution of the PET scanner could result in an attractive cost compromise for routine clinical applications.

Speaker: Paul Lecoq

PL-TBPET-Elba-2022-short.pptx

11:20 Performance evaluation of semi-monolithic detectors for TB-PET systems

The i3M is currently involved in the development of a clinical Total Body PET (TB-PET) scanner of approximately 80 cm diameter and 70 cm axial coverage. At this moment, different detectors based on semi-monolithic blocks with different surface treatments and read out by different SiPM arrays are being characterized, in order to assess how the different parameters affect the detector performance. In all cases, the external dimensions of the detector block are 25.8×25.8×20 mm3. The materials employed in the surfaces are Enhanced Specular Reflector (ESR), black painting and retroreflector. The two photodetectors employed belong to the series 13 and 14 from Hamamatsu Photonics. The results obtained for the four detectors under characterization show an energy resolution ranging from 10% up to 13% at 511 keV, a spatial and DOI resolution better than 3 mm and 4 mm, respectively, for all cases and a Detector Time Resolution (DTR) ranging from 195 ps up to 346 ps when energy-weighted averaging of the different timestamps belonging to the same event is applied.

Speaker: Marta Freire (Institute for Instrumentation in Molecular Imaging (i3m))

Freire_PSMRConference2022_SlabsTBBuena.pptx

11:40 Exploiting Cherenkov radiation and cross-luminescence emission with BGO/BaF2 metacrystals

In the field of time-of-flight positron emission tomography (PET-TOF), the time resolution of the scintillation-based detector is an essential feature. Recent studies have shown that some materials have fast emissions in the vacuum ultraviolet (VUV) region.
To acquire the fast-rising signals of these emissions, we are using optimized Coincidence Time Resolution (CTR) test boards with two output signals (timing and energy). Since the VUV silicon photomultipliers (SiPMs) do not have a protective layer on top of the metal, two different coupling materials are used and compared. The experiment is focused on Barium Fluoride (BaF2) crystals exploiting their cross-luminescence characteristic using VUV (SiPMs) from Fondazione Bruno Kessler (FBK). Moreover, a yttrium doped variant of these crystals (BaF2-Y) has been considered for comparison. The results show a DTR of 123.6 ps and 93ps for the pure and doped variants, respectively.
As reference detectors, we use a near-ultraviolet (NUV) from FBK coupled with a 3×3×5 mm3 LYSO:Ce,Ca and a 3×3×15 mm3 BGO pixels, with time resolutions (DTR) of 65.1 ps (Gaussian) and 300 ps (Laplacian), respectively.

Speaker: Riccardo Latella (Metacrystal SA)

Latella_PSMR2022_v1.5.pptx

12:00 Light Extraction Enhancement in Inorganic Scintillators for Total-body PET Scanners using Photonic Crystals

Scintillators play an important role in the detection of ionizing radiation. Improving extraction and detection of the generated light is a must to provide sufficient information on the high-energy particles interacting with the crystal. The amount of light extracted and its temporal distribution have a direct impact on the overall system performance. In positron emission tomography, energy resolution and coincidence resolving time are two of the main parameters that depend on the amount of detected light. In this study, we combine and compare several light extraction techniques for two common inorganic scintillators. An approach using a novel photonic crystal structure is also introduced. A maximum gain of ∼41% on light extraction and ∼21% on energy resolution was observed on BGO crystals using the proposed solution.

Speaker: Francesco Gramuglia (EPFL)

PSMR-TBP_LE_FG.pptx

12:20 → 15:30 Lunch break

15:30 → 18:00 Image reconstruction for PET/MR and TB-PET

Convener: Marco Aiello

15:30 MRI-guided PET Reconstruction with Adaptive Prior Strength

MRI-guided PET reconstruction can potentially reduce noise and increase spatial resolution. However, the balance between measured data and prior information usually requires manual tuning depending on the measured statistics and desired image quality. This work presents an adaptive method of MRI-guided PET reconstruction which does not require manual tuning, and is robust against a wide range of statistics and different tracer uptake distributions.

Speaker: Dr Anazodo Udunna (Lawson Health Research Institute, London, Ontario, Canada.)

psmr22_petmr_UA.pptx

15:50 AI-supported ROI list-mode reconstruction for improved lesion detectability in large-FOV PET

The extended number of detectors in large-field-of-view (LFOV) PET provides the possibility to increase the number of lines-of-response (LORs), as axial acceptance angles (α) become significantly larger when compared with conventional PET. Considering a small region inside the LFOV, however, its corresponding LORs become a very small fraction of the total detected signal and, hence, their weight within the image reconstruction (IR) strategy decreases. The dependence of signal-to-noise (SNR) and lesion detectability as a function of α are well known. This study investigates the problem further by extending a (multimodal) simulation & reconstruction management tool, named Musiré, to fully support LFOV-PET geometries and to incorporate artificial intelligence (AI)-supported methodology. Based on Monte Carlo (Gate) simulated LFOV-PET list-mode data, it was investigated whether lesion detectability can be improved by taking into account only those LORs for image reconstruction (Castor), which intersect the manually selected vicinity of a lesion within the LFOV. Having clearly observed differences in lesion contrast between global and local IR, the obvious task becomes to develop a procedure for automatic lesion classification. This has been accomplished by means of a supervised learning strategy through a convolutional neural network (U-Net/nnU-Net). Simulation design, AI-supported local lesion detection, as well as global/local list-mode reconstruction results are presented for LFOV-PET simulations. Said simulations involved anthropomorphic phantom geometries incorporating highly diverse tumor point-cloud representations. Spatial resolution and lesion-to-background contrast improved with local list-mode IR.

Speaker: Dr Joerg Peter (German Cancer Research Center)

16:10 Evaluation of different image regularization techniques on simulated phantoms with the TRIMAGE brain PET scanner

This work presents the study of the performance of the TRIMAGE brain PET scanner obtained through experimental detectors characterization, simulated phantom acquisitions and image reconstruction optimization. The TRIMAGE scanner uses dual-layer staggered LYSO:Ce crystal matrices coupled to silicon photomultipliers (SiPM). The dual layer architecture provides depth of interaction (DOI) capabilities and a finer sampling of the lines of response. Using Monte Carlo simulations, a standard-like phantom for brain imaging has been simulated. The used phantom is a smaller version of the standard image quality phantom used in whole body PET imaging, as it has been done already in similar works available in literature. Image noise and image resolution have been evaluated in terms of uniformity, recovery coefficient (RC) and spill-over ratio (SOR). In order to improve image quality, the reconstruction software has been enhanced by introducing image regularization. Different regularization algorithms have been used, including Gaussian filtering, patch-based regularization and a novel gradient-enhancer algorithm. The latter is the only iterative procedure used so far that combines a denoising filter with a feature restoration one. Image quality analysis have been performed for all the configurations mentioned above, showing that the proposed gradient minimization algorithm is very stable and performs well in terms of uniformity versus recovery coefficients, although the asymptotic values of RC and SOR are lower than in patch-based regularization.

Speaker: Luigi Masturzo (Istituto Nazionale di Fisica Nucleare)

trimage_ELBA.pdf

16:30 An unsupervised deep learning framework for respiratory motion correction in PET

Introduction: Breathing related patient motion during PET scans causes image artifacts, notably spatially variant blurring and degradation of contrast recovery. Dealing with these artifacts commonly involves respiratory gating, i.e. splitting the acquisition data into several temporal bins ("gates") depending on the respiratory cycle. With registration of the different gated images to one reference gate and subsequent averaging, motion artifacts can be minimized without increasing noise. However, traditional registration algorithms are frequently slow or of limited accuracy. In recent years, it has been shown that deep learning methods can be applied to image registration tasks. Once a neural network is trained, it can perform image registration in one pass, accelerating the process significantly. This work proposes an unsupervised deep learning framework for the registration of gated PET images.

Methods: Image pairs consisting of a fixed gate and a second moving gate serve as input to a convolutional neural network which predicts a deformation vector field (DVF) mapping the moving image to the fixed image. A spatial transformer layer is then used to warp the moving image accordingly. The network is trained in an unsupervised manner by optimizing a similarity metric between the fixed and warped images. Thus, ground truth DVFs are not required. A regularization loss is added to constrain the DVF to physically feasible motion.
Fourteen gated FDG PET/CT scans (8 gates) were available and partitioned into 11 training scans and 3 validation scans. For network training, 130 coronal slices per scan were used. With consideration of all possible gate combinations per slice that resulted in 80080 training image pairs. The normalized cross correlation (NCC) between a mid-expiration reference gate and the remaining 7 gates was calculated before and after registration as a measure of registration accuracy. The motion correction performance was evaluated for one validation scan. The motion corrected image was obtained by averaging of all gates after registration and compared to the uncorrected image, the single reference gate, and the clinically available motion correction method "OncoFreeze". Lesion SUV_max and noise levels in the liver were determined.

Results: The developed network improved the average NCC by 0.013-0.050 for all validation scans. Motion related artifacts were virtually eliminated in the investigated scan. Compared to the ungated image, the lesion SUV_max was increased (8.1 vs. 4.4, respectively) while maintaining the noise level (at 8.3%).

Conclusion: In this work, we have proposed an unsupervised registration network for respiratory motion correction in PET. Our preliminary results indicate that the framework is suitable for efficient reduction of motion related artifacts without increasing image noise compared to the uncorrected images.

Speaker: Bia Rosin (Helmholtz-Zentrum Dresden-Rossendorf, Institute of Radiopharmaceutical Cancer Research, PET Department)

psmr22rosin.pptx

16:50 Development of a deep learning method for CT-free correction for an ultra-long axial field of view PET scanner

Introduction: The possibility of low-dose positron emission tomography (PET) imaging using high sensitivity long axial field of view (FOV) PET/computed tomography (CT) scanners makes CT a critical radiation burden in clinical applications. Artificial intelligence has shown the potential to generate PET images from non-corrected PET images. Our aim in this work is to develop a CT-free correction for a long axial FOV PET scanner. Methods: Whole body PET images of 165 patients scanned with a digital regular FOV PET scanner (Biograph Vision 600 (Siemens Healthineers) in Shanghai and Bern) was included for the development and testing of the deep learning methods. Furthermore, the developed algorithm was tested on data of 7 patients scanned with a long axial FOV scanner (Biograph Vision Quadra, Siemens Healthineers). A 2D generative adversarial network (GAN) was developed featuring a residual dense block, which enables the model to fully exploit hierarchical features from all network layers. The normalized root mean squared error (NRMSE) and peak signal-to-noise ratio (PSNR), were calculated to evaluate the results generated by deep learning. Results: The preliminary results showed that, the developed deep learning method achieved an average NRMSE of 0.4±0.3% and PSNR of 51.4±6.4 for the test on Biograph Vision, and an average NRMSE of 0.5±0.4% and PSNR of 47.9±9.4 for the validation on Biograph Vision Quadra, after applied transfer learning. Conclusion: The developed deep learning method shows the potential for CT-free AI-correction for a long axial FOV PET scanner. Work in progress includes clinical assessment of PET images by independent nuclear medicine physicians. Training and fine-tuning with more datasets will be performed to further consolidate the development.

Speaker: Song Xue

Song_Xue.pptx

17:10 Invited Talk - Deep learning reconstruction for PET-MR and total-body PET: present status and future perspectives

Speaker: Andrew Reader (Kings College London)

AJReader_Slidesv2.pdf

17:50 Final remarks

final_remarks.pdf

18:00 → 19:00 Farewell cocktail