Holistic evaluation of a machine learning-based timing calibration for PET detectors under varying data sparsity

Objective.Modern PET scanners offer precise TOF information, improving the SNR of the reconstructed images. Timing calibrations are performed to reduce the worsening effects of the system components and provide valuable TOF information. Traditional calibration procedures often provide static or linear corrections, with the drawback that higher-order skews or event-to-event corrections are not addressed. Novel research demonstrated significant improvements in the reachable timing resolutions when combining conventional calibration approaches with machine learning, with the disadvantage of extensive calibration times infeasible for a clinical application. In this work, we made the first steps towards an in-system application and analyzed the effects of varying data sparsity on a machine learning timing calibration, aiming to accelerate the calibration time. Furthermore, we demonstrated the versatility of our calibration concept by applying the procedure for the first time to analog readout technology.Approach.We modified experimentally acquired calibration data used for training regarding their statistical and spatial sparsity, mimicking reduced measurement time and variability of the training data. Trained models were tested on unseen test data, characterized by fine spatial sampling and rich statistics. In total, 80 decision tree models with the same hyperparameter settings, were trained and holistically evaluated regarding data scientific, physics-based, and PET-based quality criteria.Main results.The calibration procedure can be heavily reduced from several days to some minutes without sacrificing quality and still significantly improving the timing resolution from(304±5)psto(216±1)pscompared to conventionally used analytical calibration methods.Significance.This work serves as the first step in making the developed machine learning-based calibration suitable for an in-system application to profit from the method's capabilities on the system level. Furthermore, this work demonstrates the functionality of the methodology on detectors using analog readout technology. The proposed holistic evaluation criteria here serve as a guideline for future evaluations of machine learning-based calibration approaches.

A finely segmented semi-monolithic detector tailored for high-resolution PET

BACKGROUND

Preclinical research and organ-dedicated applications use and require high (spatial-)resolution positron emission tomography (PET) detectors to visualize small structures (early) and understand biological processes at a finer level of detail. Researchers seeking to improve detector and image spatial resolution have explored various detector designs. Current commercial high-resolution systems often employ finely pixelated or monolithic scintillators, each with its limitations.

PURPOSE

We present a semi-monolithic detector, tailored for high-resolution PET applications with a spatial resolution in the range of 1 mm or better, merging concepts of monolithic and pixelated crystals. The detector features LYSO slabs measuring (24 × 10 × 1) mm3, coupled to a 12 × 12 readout channel photosensor with 4 mm pitch. The slabs are grouped in two arrays of 44 slabs each to achieve a higher optical photon density despite the fine segmentation.

METHODS

We employ a fan beam collimator for fast calibration to train machine-learning-based positioning models for all three dimensions, including slab identification and depth-of-interaction (DOI), utilizing gradient tree boosting (GTB). The data for all dimensions was acquired in less than 2 h. Energy calculation was based on a position-dependent energy calibration. Using an analytical timing calibration, time skews were corrected for coincidence timing resolution (CTR) estimation.

RESULTS

Leveraging machine-learning-based calibration in all three dimensions, we achieved high detector spatial resolution: down to 1.18 mm full width at half maximum (FWHM) detector spatial resolution and 0.75 mm mean absolute error (MAE) in the planar-monolithic direction, and 2.14 mm FWHM and 1.03 mm MAE for DOI at an energy window of (435-585) keV. Correct slab interaction identification in planar-segmented direction exceeded 80%, alongside an energy resolution of 12.7% and a CTR of 450 ps FWHM.

CONCLUSIONS

The introduced finely segmented, high-resolution slab detector demonstrates appealing performance characteristics suitable for high-resolution PET applications. The current benchtop-based detector calibration routine allows these detectors to be used in PET systems.

DOI- and TOF-capable PET array detector using double-ended light readout and stripline-based row and column electronic readout

We investigate a highly multiplexing readout for depth-of-interaction (DOI) and time-of-flight PET detector consisting of an N×N crystals whose light outputs at the front and back ends are detected by using silicon photomultipliers (SiPM). The front N×N SiPM array is read by using a stripline (SL) configured to support discrimination of the row position of the signal-producing crystal. The back N×N SiPM array is similarly read by an SL for column discrimination. Hence, the detector has only four outputs. We built 4×4 and 8×8 detector modules (DM) by using 3.0×3.0×20 mm3 lutetium-yttrium oxyorthosilicates. The outputs were sampled and processed offline. For both DMs, crystal discrimination was successful. For the 4×4 DM, we obtained an average energy resolution (ER) of 14.1%, an average DOI resolution of 2.5 mm, a non DOI-corrected coincidence resolving time (CRT), measured in coincidence with a single-pixel reference detector, of about 495 ps. For the 8×8 DM, the average ER, average DOI resolution and average CRT were 16.4%, 2.9 mm, and 641 ps, respectively. We identified the intercrystal scattering as a probable cause for the CRT deterioration when the DM was increased from 4×4 to 8×8.

Advancements in Positron Emission Tomography Detectors: From Silicon Photomultiplier Technology to Artificial Intelligence Applications

This review article focuses on PET detector technology, which is the most crucial factor in determining PET image quality. The article highlights the desired properties of PET detectors, including high detection efficiency, spatial resolution, energy resolution, and timing resolution. Recent advancements in PET detectors to improve these properties are also discussed, including the use of silicon photomultiplier technology, advancements in depth-of-interaction and time-of-flight PET detectors, and the use of artificial intelligence for detector development. The article provides an overview of PET detector technology and its recent advancements, which can significantly enhance PET image quality.

Near-field coded-mask technique and its potential for proton therapy monitoring

Objective.Prompt-gamma imaging encompasses several approaches to the online monitoring of the beam range or deposited dose distribution in proton therapy. We test one of the imaging techniques - a coded mask approach - both experimentally and via simulations.Approach.Two imaging setups have been investigated experimentally. Each of them comprised a structured tungsten collimator in the form of a modified uniformly redundant array mask and a LYSO:Ce scintillation detector of fine granularity. The setups differed in detector dimensions and operation mode (1D or 2D imaging). A series of measurements with radioactive sources have been conducted, testing the performance of the setups for near-field gamma imaging. Additionally, Monte Carlo simulations of a larger setup of the same type were conducted, investigating its performance with a realistic gamma source distribution occurring during proton therapy.Main results.The images of point-like sources reconstructed from two small-scale prototypes' data using the maximum-likelihood expectation maximisation algorithm constitute the experimental proof of principle for the near-field coded-mask imaging modality, both in the 1D and the 2D mode. Their precision allowed us to calibrate out certain systematic offsets appearing due to the limited alignment accuracy of setup elements. The simulation of the full-scale setup yielded a mean distal falloff retrieval precision of 0.72 mm in the studies for beam energy range 89.5-107.9 MeV and with 1 × 108protons (a typical number for distal spots). The implemented algorithm of image reconstruction is relatively fast-a typical procedure needs several seconds.Significance.Coded-mask imaging appears a valid option for proton therapy monitoring. The results of simulations let us conclude that the proposed full-scale setup is competitive with the knife-edge-shaped and the multi-parallel slit cameras investigated by other groups.

Applications of machine learning and deep learning in SPECT and PET imaging: General overview, challenges and future prospects

The integration of positron emission tomography (PET) and single-photon emission computed tomography (SPECT) imaging techniques with machine learning (ML) algorithms, including deep learning (DL) models, is a promising approach. This integration enhances the precision and efficiency of current diagnostic and treatment strategies while offering invaluable insights into disease mechanisms. In this comprehensive review, we delve into the transformative impact of ML and DL in this domain. Firstly, a brief analysis is provided of how these algorithms have evolved and which are the most widely applied in this domain. Their different potential applications in nuclear imaging are then discussed, such as optimization of image adquisition or reconstruction, biomarkers identification, multimodal fusion and the development of diagnostic, prognostic, and disease progression evaluation systems. This is because they are able to analyse complex patterns and relationships within imaging data, as well as extracting quantitative and objective measures. Furthermore, we discuss the challenges in implementation, such as data standardization and limited sample sizes, and explore the clinical opportunities and future horizons, including data augmentation and explainable AI. Together, these factors are propelling the continuous advancement of more robust, transparent, and reliable systems.

Perspectives of Evidence-Based Therapy Management

BACKGROUND

Therapeutics that specifically address biological processes often require a much finer selection of patients and subclassification of diseases. Thus, diagnostic procedures must describe the diseases in sufficient detail to allow selection of appropriate therapy and to sensitively track therapy response. Anatomical features are often not sufficient for this purpose and there is a need to image molecular and pathophysiological processes.

METHOD

Two imaging strategies can be pursued: molecular imaging attempts to image a few biomarkers that play key roles in pathological processes. Alternatively, patterns describing a biological process can be identified from the synopsis of multiple (non-specific) imaging markers, possibly in combination with omics and other clinical findings. Here, AI-based methods are increasingly being used.

RESULTS

Both strategies of evidence-based therapy management are explained in this review article and examples and clinical successes are presented. In this context, reviews of clinically approved molecular diagnostics and decision support systems are listed. Furthermore, since reliable, representative, and sufficiently large datasets are further important prerequisites for AI-assisted multiparametric analyses, concepts are presented to make data available in a structured way, e. g., using Generative Adversarial Networks to complement databases with virtual cases and to build completely anonymous reference databases.

CONCLUSION

Molecular imaging and computer-assisted cluster analysis of diagnostic data are complementary methods to describe pathophysiological processes. Both methods have the potential to improve (evidence-based) the future management of therapies, partly on their own but also in combined approaches.

KEY POINTS

· Molecular imaging and radiomics provide valuable complementary disease biomarkers.. · Data-driven, model-based, and hybrid model-based integrated diagnostics advance precision medicine.. · Synthetic data generation may become essential in the development process of future AI methods..

Timing, Energy, and Spatial Characterization of Highly Sampled Monolithic PET Detectors with Different Thicknesses

The HyPET project proposes a hybrid dedicated TOF-PET for prostate imaging, with pixelated detector blocks in the front layer and monolithic blocks in the back layer. In this work, four detector configurations have been experimentally evaluated for the rear detector layer. The detector configuration consists of LYSO monolithic blocks with the same size (25.4 mm × 25.4 mm) but different thicknesses (5, 7.5, 10, and 15 mm) coupled to the same SiPM array. Each detector configuration has been experimentally characterized in terms of time, energy and spatial resolution by scanning the crystal surface using a fan beam in steps of 0.25 mm. Regarding spatial resolution, the interaction position was estimated using a Neural Network technique. All resolutions except energy, which remains nearly constant at 17% for all cases, show better values for the 5 mm detector thickness. We have achieved spatial resolution values of FWHM of 1.02 ± 0.10, 1.19 ± 0.13, 1.53 ± 0.17, 2.33 ± 0.55 mm, for the 5, 7.5, 10, and 15 mm blocks, respectively. The detector time resolution obtained was 275 ± 26, 291 ± 21, 344 ± 48, and 433 ± 45 ps respectively, using the energy weighted average method for the time stamps.

Resolution estimation in different monolithic PET detectors using neural networks

PURPOSE

We use neural networks to evaluate and compare the spatial resolution of two different simulated monolithic PET detector elements. The effects of mixing events with single photoeffect interactions and multiple Compton scatterings are also studied.

METHODS

Two PET detector models were used in this study. The first one consisted of a LYSO crystal plate with 19.25 × 19.25 × 12 mm³ dimensions and 256-channel photomultiplier with parameters modeled after a Hamamatsu S-13615-1050N-16 SiPM. The second model used a larger LYSO crystal (57.6 × 57.6 × 12 mm³) and a 64-channel Sensl ARRAYC-60035-64P-PCB photomultiplier. A feed-forward neural network was used to reconstruct the point of 511 keV gamma interaction. The number of layers and the number of neurons per layer were varied.

RESULTS

The best resolution was achieved with the 57.6 × 57.6 mm² detector model, with an average of 0.74 ± 0.01 mm for the XY plane and an average 1.01 ± 0.01 mm for the Z coordinate (depth of interaction).

CONCLUSIONS

Neural networks can be a powerful tool that can help to determine the optimal parameters for a design of an experimental device. This study demonstrates how neural networks can be used to evaluate the performance of two detector variants while not being dependent on specific signal and noise functions.

Analysis of a convex time skew calibration for light sharing-based PET detectors

Objective.Positron emission tomography (PET) detectors providing attractive coincidence time resolutions (CTRs) offer time-of-flight information, resulting in an improved signal-to-noise ratio of the PET image. In applications with photosensor arrays that employ timestampers for individual channels, timestamps typically are not time synchronized, introducing time skews due to different signal pathways. The scintillator topology and transportation of the scintillation light might provoke further skews. If not accounted for these effects, the achievable CTR deteriorates. We studied a convex timing calibration based on a matrix equation. In this work, we extended the calibration concept to arbitrary structures targeting different aspects of the time skews and focusing on optimizing the CTR performance for detector characterization. The radiation source distribution, the stability of the estimations, and the energy dependence of calibration data are subject to the analysis.Approach.A coincidence setup, equipped with a semi-monolithic detector comprising 8 LYSO slabs, each 3.9 mm × 31.9 mm × 19.0 mm, and a one-to-one coupled detector with 8 × 8 LYSO segments of 3.9 mm × 3.9 mm × 19.0 mm volume is used. Both scintillators utilize a dSiPM (DPC3200-22-44, Philips Digital Photon Counting) operated in first photon trigger. The calibration was also conducted with solely one-to-one coupled detectors and extrapolated for a slab-only setup.Main results.All analyzed hyperparameters show a strong influence on the calibration. Using multiple radiation positions improved the skew estimation. The statistical significance of the calibration dataset and the utilized energy window was of great importance. Compared to a one-to-one coupled detector pair achieving CTRs of 224 ps the slab detector configuration reached CTRs down to 222 ps, demonstrating that slabs can compete with a clinically used segmented detector design.Significance.This is the first work that systematically studies the influence of hyperparameters on skew estimation and proposes an extension to arbitrary calibration structures (e.g. scintillator volumes) of a known calibration technique.

Timing, Energy, and 3-D Spatial Resolution of the BING PET Detector Module

We evaluated the 3D spatial, energy, and timing resolution of the Brain (or Breast)-Initiative Next-Generation (BING) PET detector. The BING detector is an array of 1-mm-thick slats of LYSO scintillator with lapped specular-reflective faces (15-mm by 52-mm) that are stacked together and oriented with their long-narrow edges normal to the imaging field of view. Interaction positions are determined from the signals of silicon-photomultiplier (SiPM) arrays placed on the entrance (top) and exit (bottom) faces. The SiPM arrays are offset to determine the slat of interaction (SOI) without requiring any optical light sharing between slats. Maximum likelihood 2D location within the SOI is determined using the sensor signals. Interaction time is determined with a modified first-optical-photon pickoff method. Performance of the BING detector was measured as a function of position using a sideways coincidence-collimated beam. Slats were accurately identified, with an effective tangential detector resolution of 1 mm. Average resolutions (and ranges) are: 0.96 mm (0.85 mm to 1.11 mm) for lateral (axial) detector resolution, 1.6 mm (1.0 mm to 2.1 mm) for depth resolution, 13.6% (12.7% to 16.0%) for energy resolution, and 317 ps (241 ps to 404 ps) for coincidence timing resolution. Initial spatial and timing resolution results demonstrated that the BING detector can be effective in a small field-of view (e.g., brain or breast) PET system.

Position estimation using neural networks in semi-monolithic PET detectors

Objective. The goal of this work is to experimentally compare the 3D spatial and energy resolution of a semi-monolithic detector suitable for total-body positron emission tomography (TB-PET) scanners using different surface crystal treatments and silicon photomultiplier (SiPM) models.Approach. An array of 1 × 8 lutetium yttrium oxyorthosilicate (LYSO) slabs of 25.8 × 3.1 × 20 mm³separated with Enhanced Specular Reflector (ESR) was coupled to an array of 8 × 8 SiPMs. Three different treatments for the crystal were evaluated: ESR + RR + B,with lateral faces black (B) painted and a retroreflector (RR) layer added to the top face; ESR +RR, with lateral faces covered with ESR and a RR layer on the top face and; All ESR, with lateral and top sides with ESR. Additionally, two SiPM array models from Hamamatsu Photonics belonging to the series S13361-3050AE-08 (S13) and S14161-3050AS-08 (S14) have been compared. Coincidence data was experimentally acquired using a²²Na point source, a pinhole collimator, a reference detector and moving the detector under study in 1 mm steps in thex- andDOI- directions. The spatial performance was evaluated by implementing a neural network (NN) technique for the impact position estimation in thex- (monolithic) andDOIdirections.Results. Energy resolution values of 16 ± 1%, 11 ± 1%, 16 ± 1%, 15 ± 1%, and 13 ± 1% were obtained for theS13-ESR + B + RR,S13-AllESR,S14-ESR + B + RR,S14-ESR + RR,andS14-AllESR, respectively. Regarding positioning accuracy, mean average error of 1.1 ± 0.5, 1.3 ± 0.5 and 1.3 ± 0.5 were estimated for thex- direction and 1.7 ± 0.8, 2.0 ± 0.9 and 2.2 ± 1.0 for theDOI- direction, for the ESR + B + RR, ESR + RR and All ESR cases, respectively, regardless of the SiPM model.Significance. Overall, the obtained results show that the proposed semi-monolithic detectors are good candidates for building TB-PET scanners.

A semi-monolithic detector providing intrinsic DOI-encoding and sub-200 ps CRT TOF-capabilities for clinical PET applications

BACKGROUND

Current clinical positron emission tomography (PET) systems utilize detectors where the scintillator typically contains single elements of 3-6-mm width and about 20-mm height. While providing good time-of-flight performance, this design limits the spatial resolution and causes radial astigmatism as the depth-of-interaction (DOI) remains unknown.

PURPOSE

We propose an alternative, aiming to combine the advantages of current detectors with the DOI capabilities shown for monolithic concepts, based on semi-monolithic scintillators (slabs). Here, the optical photons spread along one dimension enabling DOI-encoding with a still small readout area beneficial for timing performance.

METHODS

An array of eight monolithic LYSO slabs of dimensions 3.9 × 32 × 19 mm³ was read out by a 64-channel photosensor containing digital SiPMs (DPC3200-22-44, Philips Digital Photon Counting). The position estimation in the detector's monolithic and DOI direction was based on a calibration with a fan beam collimator and the machine learning technique gradient tree boosting (GTB).

RESULTS

We achieved a positioning performance in terms of mean absolute error (MAE) of 1.44 mm for the monolithic direction and 2.12 mm for DOI considering a wide energy window of 300-700 keV. The energy resolution was determined to be 11.3%, applying a positional-dependent energy calibration. We established both an analytical and machine-learning-based timing calibration approach and applied them for a first-photon trigger. The analytical timing calibration corrects for electronic and optical time skews leading to 240 ps coincidence resolving time (CRT) for a pair of slab-detectors. The CRT was significantly improved by utilizing GTB to predict the time difference based on specific training data and applied on top of the analytical calibration. We achieved 209 ps for the wide energy window and 198 ps for a narrow selection around the photopeak (411-561 keV). To maintain the detector's sensitivity, no filters were applied to the data during processing.

CONCLUSION

Overall, the semi-monolithic detector provides attractive performance characteristics. Especially, a good CRT can be achieved while introducing DOI capabilities to the detector, making the concept suitable for clinical PET scanners.

Perspectives of Evidence-Based Therapy Management

BACKGROUND

 Therapeutics that specifically address biological processes often require a much finer selection of patients and subclassification of diseases. Thus, diagnostic procedures must describe the diseases in sufficient detail to allow selection of appropriate therapy and to sensitively track therapy response. Anatomical features are often not sufficient for this purpose and there is a need to image molecular and pathophysiological processes.

METHOD

 Two imaging strategies can be pursued: molecular imaging attempts to image a few biomarkers that play key roles in pathological processes. Alternatively, patterns describing a biological process can be identified from the synopsis of multiple (non-specific) imaging markers, possibly in combination with omics and other clinical findings. Here, AI-based methods are increasingly being used.

RESULTS

 Both strategies of evidence-based therapy management are explained in this review article and examples and clinical successes are presented. In this context, reviews of clinically approved molecular diagnostics and decision support systems are listed. Furthermore, since reliable, representative, and sufficiently large datasets are further important prerequisites for AI-assisted multiparametric analyses, concepts are presented to make data available in a structured way, e. g., using Generative Adversarial Networks to complement databases with virtual cases and to build completely anonymous reference databases.

CONCLUSION

 Molecular imaging and computer-assisted cluster analysis of diagnostic data are complementary methods to describe pathophysiological processes. Both methods have the potential to improve (evidence-based) the future management of therapies, partly on their own but also in combined approaches.

KEY POINTS

· Molecular imaging and radiomics provide valuable complementary disease biomarkers.. · Data-driven, model-based, and hybrid model-based integrated diagnostics advance precision medicine.. · Synthetic data generation may become essential in the development process of future AI methods..

CITATION FORMAT

· Kiessling F, Schulz V, . Perspectives of Evidence-Based Therapy Management. Fortschr Röntgenstr 2022; DOI: 10.1055/a-1752-0839.

Nuclear Medicine and Artificial Intelligence: Best Practices for Algorithm Development

The nuclear medicine field has seen a rapid expansion of academic and commercial interest in developing artificial intelligence (AI) algorithms. Users and developers can avoid some of the pitfalls of AI by recognizing and following best practices in AI algorithm development. In this article, recommendations on technical best practices for developing AI algorithms in nuclear medicine are provided, beginning with general recommendations and then continuing with descriptions of how one might practice these principles for specific topics within nuclear medicine. This report was produced by the AI Task Force of the Society of Nuclear Medicine and Molecular Imaging.

Hybrid total-body pet scanners-current status and future perspectives

Purpose Since the 1990s, PET has been successfully combined with MR or CT systems. In the past years, especially PET systems have seen a trend towards an enlarged axial field of view (FOV), up to a factor of ten. Methods Conducting a thorough literature research, we summarize the status quo of contemporary total-body (TB) PET/CT scanners and give an outlook on possible future developments. Results Currently, three human TB PET/CT systems have been developed: The PennPET Explorer, the uExplorer, and the Biograph Vision Quadra realize aFOVs between 1 and 2 m and show a tremendous increase in system sensitivity related to their longer gantries. Conclusion The increased system sensitivity paves the way for short-term, low-dose, and dynamic TB imaging as well as new examination methods in almost all areas of imaging.

Artificial Intelligence and Cardiac PET/Computed Tomography Imaging

Artificial intelligence is an important technology, with rapidly expanding applications for cardiac PET. We review the common terminology, including methods for training and testing, which are fundamental to understanding artificial intelligence. Next, we highlight applications to improve image acquisition, reconstruction, and segmentation. Computed tomographic imaging is commonly acquired in conjunction with PET and various artificial intelligence methods have been applied, including methods to automatically extract anatomic information or generate synthetic attenuation images. Last, we describe methods to automate disease diagnosis or risk stratification. This summary highlights the current and future clinical applications of artificial intelligence to cardiovascular PET imaging.

Development of dual-ended depth-of-interaction detectors using laser-induced crystals for small animal PET systems

Sensitivity and spatial resolution of positron emission tomography (PET) scanners can be improved by using thicker scintillation crystals with depth-of-interaction (DOI) encoding. Subsurface laser engraving (SSLE) can be used to segment crystals of a scintillation detector in order to fabricate a DOI detector. We previously applied SSLE to crystal bars of 3 × 3 × 20 mm³and 1.5 × 1.5 × 20 mm³and developed two dual-ended detectors with DOI segments of 3 mm and 1.5 mm, respectively. To further improve the DOI resolution, our SSLE detector design can be used with smaller pitch crystal bars, making them excellent detector candidates for small animal PET scanners with submillimetre resolution. In the present study, three small crystal bars of 1 × 1 × 20 mm³, 2 × 1 × 20 mm³, and 2 × 1 × 40 mm³were laser engraved to 12, 20 and 40 segments, respectively, by applying SSLE in their height directions. The segmented crystal bars were characterised in three prototype detector arrangements. First, the 1 × 1 × 20 mm³crystal bars were characterised in an 8 × 8 crystal array designed for DOI encoding along crystal height in a conventional small animal PET design. Second, a 4 × 8 crystal array of 2 × 1 × 20 mm³crystal bars was characterised for using the DOI information for crystal interaction positioning along the axial axis of a small animal PET scanner. Finally, the third part of the study was performed on a single 2 × 1 × 40 mm³crystal bar with 40 segments to investigate the feasibility of DOI estimation in longer crystals for application in a system with extended axial length. We evaluated the capability of segment identification and energy resolution of theses detectors. The 3D position maps of the detectors were obtained using the Anger-type calculation and the crystal identification performance was evaluated for each detector. Clear segment separation was obtained for the crystal arrays with 12 (segment pitch of 1.67 mm) and 20 (segment pitch of 1 mm) segments. Mean energy resolutions of 8.8% ± 0.4% and 9.6% ± 0.8% at 511 keV were obtained for the segments in the central regions of the 8 × 8 array with 12 segments and the 4 × 8 array with 20 segments, respectively. Clear segment identification was found to be difficult for the detector with 40 segments, especially for the segments at the middle of the crystal. Energy and interaction positioning characterisation results suggest that both prototype detectors with 12 and 20 segments are well suited for small animal PET scanners with high spatial resolution.

High throughput software-based gradient tree boosting positioning for PET systems

The supervised machine learning technique Gradient Tree Boosting (GTB) has shown good accuracy for position estimation of gamma interaction in PET crystals for bench-top experiments while its computational requirements can easily be adjusted. Transitioning to preclinical and clinical applications requires near real-time processing in the scale of full PET systems. In this work, a high throughput GTB-based singles positioning C++ implementation is proposed and a series of optimizations are evaluated regarding their effect on the achievable processing throughput. Moreover, the crucial feature and parameter selection for GTB is investigated for the segmented detectors of the Hyperion II^DPET insert with two main models and a range of GTB hyperparameters. The proposed framework achieves singles positioning throughputs of more than 9.5 GB/s for smaller models and of 240 MB/s for more complex models on a recent Intel Skylake server. Detailed throughput analysis reveals the key performance limiting factors, and an empirical throughput model is derived to guide the GTB model selection process and scanner design decisions. The throughput model allows for throughput estimations with a mean absolute error (MAE) of 175.78 MB/s.

Advanced Monte Carlo simulations of emission tomography imaging systems with GATE

Built on top of the Geant4 toolkit, GATE is collaboratively developed for more than 15 years to design Monte Carlo simulations of nuclear-based imaging systems. It is, in particular, used by researchers and industrials to design, optimize, understand and create innovative emission tomography systems. In this paper, we reviewed the recent developments that have been proposed to simulate modern detectors and provide a comprehensive report on imaging systems that have been simulated and evaluated in GATE. Additionally, some methodological developments that are not specific for imaging but that can improve detector modeling and provide computation time gains, such as Variance Reduction Techniques and Artificial Intelligence integration, are described and discussed.

DOI estimation through signal arrival time distribution: a theoretical description including proof of concept measurements

The challenge to reach 10 ps coincidence time resolution (CTR) in time-of-flight positron emission tomography (TOF-PET) is triggering major efforts worldwide, but timing improvements of scintillation detectors will remain elusive without depth-of-interaction (DOI) correction in long crystals. Nonetheless, this momentum opportunely brings up the prospect of a fully time-based DOI estimation since fast timing signals intrinsically carry DOI information, even with a traditional single-ended readout. Consequently, extracting features of the detected signal time distribution could uncover the spatial origin of the interaction and in return, provide enhancement on the timing precision of detectors. We demonstrate the validity of a time-based DOI estimation concept in two steps. First, experimental measurements were carried out with current LSO:Ce:Ca crystals coupled to FBK NUV-HD SiPMs read out by fast high-frequency electronics to provide new evidence of a distinct DOI effect on CTR not observable before with slower electronics. Using this detector, a DOI discrimination using a double-threshold scheme on the analog timing signal together with the signal intensity information was also developed without any complex readout or detector modification. As a second step, we explored by simulation the anticipated performance requirements of future detectors to efficiently estimate the DOI and we proposed four estimators that exploit either more generic or more precise features of the DOI-dependent timestamp distribution. A simple estimator using the time difference between two timestamps provided enhanced CTR. Additional improvements were achieved with estimators using multiple timestamps (e.g. kernel density estimation and neural network) converging to the Cramér-Rao lower bound developed in this work for a time-based DOI estimation. This two-step study provides insights on current and future possibilities in exploiting the timing signal features for DOI estimation aiming at ultra-fast CTR while maintaining detection efficiency for TOF PET.

Evaluation of the PETsys TOFPET2 ASIC in multi-channel coincidence experiments

BACKGROUND

Aiming to measure the difference in arrival times of two coincident γ-photons with an accuracy in the order of 200ps, time-of-flight positron emission tomography systems commonly employ silicon photomultipliers (SiPMs) and high-resolution digitization electronics, application specific integrated circuits (ASICs). This work evaluates the performance of the TOFPET2 ASIC, released by PETsys Electronics S.A. in 2017, dependent on its configuration parameters in multi-channel coincidence measurements.

METHODS

SiPM arrays fabricated by different vendors (KETEK, SensL, Hamamatsu, Broadcom) were tested in combination with the ASIC. Scintillator arrays featuring different reflector designs and different configurations of the TOFPET2 ASIC software parameters were evaluated. The benchtop setup used is provided with the TOFPET2 ASIC evaluation kit by PETsys Electronics S.A.

RESULTS

Compared to existing studies featuring the TOFPET2 ASIC, multi-channel performance results dependent on a larger set of ASIC configuration parameters were obtained that have not been reported to this extend so far. The ASIC shows promising CRTs down to 219.9 ps in combination with two Hamamatsu S14161-3050-HS-08 SiPM arrays (128 channels read out, energy resolution 13.08%) and 216.1 ps in combination with two Broadcom AFBR-S4N44P643S SiPM arrays (32 channels read out, energy resolution 9.46%). The length of the trigger delay of the dark count suppression scheme has an impact on the ASIC performance and can be configured to further improve the coincidence resolution time. The integrator gain configuration has been investigated and allows an absolute improvement of the energy resolution by up to 1% at the cost of the linearity of the energy spectrum.

CONCLUSION

Measuring up to the time-of-flight performance of state-of-the-art positron emission tomography (ToF-PET) systems while providing a uniform and stable readout for multiple channels at the same time, the TOFPET2 ASIC is treated as promising candidate for the integration in future ToF-PET systems.

Artificial neural networks for positioning of gamma interactions in monolithic PET detectors

To detect gamma rays with good spatial, timing and energy resolution while maintaining high sensitivity we need accurate and efficient algorithms to estimate the first gamma interaction position from the measured light distribution. Furthermore, monolithic detectors are investigated as an alternative to pixelated detectors due to increased sensitivity, resolution and intrinsic DOI encoding. Monolithic detectors, however, are challenging because of complicated calibration setups and edge effects. In this work, we evaluate the use of neural networks to estimate the 3D first (Compton or photoelectric) interaction position. Using optical simulation data of a 50×50×16 mm³LYSO crystal, performance is evaluated as a function of network complexity (two to five hidden layers with 64 to 1024 neurons) and amount of training data (1000 to 8000 training events per calibration position). We identify and address the potential pitfall of overfitting on the training grid through evaluation on intermediate positions that are not in the training set. Additionally, the performance of neural networks is directly compared with nearest neighbour positioning. Optimal performance was achieved with a network containing three hidden layers of 256 neurons trained on 1000 events/position. For more complex networks, the performance degrades at intermediate positions and overfitting starts to occur. A median 3D positioning error of 0.77 mm and a 2D FWHM of 0.46 mm is obtained. This is a 17% improvement in terms of FWHM compared to the nearest neighbour algorithm. Evaluation only on events that are not Compton scattered results in a 3D positioning error of 0.40 mm and 2D FWHM of 0.42 mm. This reveals that Compton scatter results in a considerable increase of 93% in positioning error. This study demonstrates that very good spatial resolutions can be achieved with neural networks, superior to nearest neighbour positioning. However, potential overfitting on the training grid should be carefully evaluated.

SiPM signal readout for inter-crystal scatter event identification in PET detectors

In positron emission tomography (PET) with pixelated detectors, a significant number of annihilation photons interact with scintillation crystals through single or multiple Compton scattering events. When these partial energy depositions occur across multiple crystal elements, we call them inter-crystal scatter (ICS) events. ICS events lead to incorrect localization of the annihilation photons, thereby degrading the PET image contrast, spatial resolution, and lesion detectability. The accurate identification of ICS events is the first essential step to improve the quality of PET images by rejecting ICS events or recovering ICS events without losing PET sensitivity. In this study, we propose a novel silicon photomultiplier (SiPM) readout method to identify ICS events in one-to-one coupled PET detectors with a reduced number of data acquisition channels. For concept verification, we assembled a PET detector that consists of a 16-channel SiPM array and 4 [Formula: see text] 4 lutetium oxyorthosilicate (LSO) array with a 3.2 mm crystal pitch. The proposed SiPM readout scheme serializes the 16 SiPM anode signals into four pulse train outputs encoded with four increasing time-delays in steps of 250 ns intervals. A Sum signal of the 16 SiPM anodes provides the timing information for time-of-flight measurement and a trigger signal for coincidence detection. A time-over-threshold (TOT) method was applied for obtaining the energy information followed by a subsequent TOT-to-energy calibration. We successfully identified the ICS events and determined their interacted positions and deposited energies by analyzing the digital pulses from the four pulse train output channels. The occurrence rate of ICS events was 10.85% for the 4 × 4 PET detector module with 3.2 mm-pitch LSO crystals. The PET detector yielded an energy resolution of 10.9 [Formula: see text] 0.6% and coincidence timing resolution of 285 [Formula: see text] 12 ps FWHM. We expect that the proposed method can be a useful solution for alleviating the readout burden of SiPM-based PET scanners with ICS event identification capability.

Performance comparison of dual-ended readout depth-encoding PET detectors based on BGO and LYSO crystals

The performance of dual-ended readout depth-encoding PET detectors based on bismuth germanate (BGO) coupled to silicon photomultipliers (SiPM) arrays was measured for the first time and compared to lutetium-yttrium oxyorthosilicate (LYSO) based detectors using the same readout. The BGO and LYSO crystal arrays all had a crystal pitch of 2.2 mm and were coupled to 8 × 8 SiPM arrays with a matching pitch of 2.2 mm, using a one-to-one coupling configuration. Three types of crystals with Toray reflector were used: polished LYSO, polished BGO, and unpolished BGO, and for two different crystal thicknesses of 20 mm and 30 mm. All the crystal elements in the BGO arrays were clearly resolved in the flood histogram. Better flood histograms were obtained using the LYSO arrays for a selected crystal thickness, and better flood histograms were obtained using the 20 mm thick crystal arrays for a selected crystal type. The average crystal level energy resolution and timing resolution for 20 mm polished LYSO, polished BGO and unpolished BGO crystals at their optimal SiPM bias voltage were 18.6 ± 1.3 % and 1.19 ± 0.20 ns, 17.8 ± 0.8 % and 4.43 ± 0.47 ns, and 18.0 ± 1.0 % and 4.68 ± 1.0 ns, respectively. Depth-of-interaction (DOI) resolution of the 20 mm polished LYSO array was 2.31 ± 0.17 mm and for the 20 mm unpolished BGO array was 3.53 ± 0.25 mm. However, polished BGO arrays with Toray reflector did not provide DOI information. Our key conclusion is that dual-ended readout depth-encoding 20 mm thick unpolished BGO detectors are good candidates for low-activity PET systems with small field-of-view and low timing performance requirements, such as preclinical or compact organ-dedicated PET systems, with the advantage over LYSO of having no background radiation and significantly lower cost.

Applications of artificial intelligence and deep learning in molecular imaging and radiotherapy

This brief review summarizes the major applications of artificial intelligence (AI), in particular deep learning approaches, in molecular imaging and radiation therapy research. To this end, the applications of artificial intelligence in five generic fields of molecular imaging and radiation therapy, including PET instrumentation design, PET image reconstruction quantification and segmentation, image denoising (low-dose imaging), radiation dosimetry and computer-aided diagnosis, and outcome prediction are discussed. This review sets out to cover briefly the fundamental concepts of AI and deep learning followed by a presentation of seminal achievements and the challenges facing their adoption in clinical setting.

Improving depth-of-interaction resolution in pixellated PET detectors using neural networks

Parallax error is a common issue in high-resolution preclinical positron emission tomography (PET) scanners as well as in clinical scanners that have a long axial field of view (FOV), which increases estimation uncertainty of the annihilation position and therefore degrades the spatial resolution. A way to address this issue is depth-of-interaction (DOI) estimation. In this work we propose two machine learning-based algorithms, a dense and a convolutional neural network (NN), as well as a multiple linear regression (MLR)-based method to estimate DOI in depolished PET detector arrays with single-sided readout. The algorithms were tested on an 8× 8 array of 1.53× 1.53× 15 mm³ crystals and a 4× 4 array of 3.1× 3.1× 15 mm³ crystals, both made of Ce:LYSO scintillators and coupled to a 4× 4 array of 3× 3 mm³ silicon photomultipliers (SiPMs). Using the conventional linear DOI estimation method resulted in an average DOI resolution of 3.76 mm and 3.51 mm FWHM for the 8× 8 and the 4× 4 arrays, respectively. Application of MLR outperformed the conventional method with average DOI resolutions of 3.25 mm and 3.33 mm FWHM, respectively. Using the machine learning approaches further improved the DOI resolution, to an average DOI resolution of 2.99 mm and 3.14 mm FWHM, respectively, and additionally improved the uniformity of the DOI resolution in both arrays. Lastly, preliminary results obtained by using only a section of the crystal array for training showed that the NN-based methods could be used to reduce the number of calibration steps required for each detector array.

Depth of Interaction Estimation in a Preclinical PET Scanner Equipped with Monolithic Crystals Coupled to SiPMs Using a Deep Neural Network

The scintillation light distribution produced by photodetectors in positron emission tomography (PET) provides the depth of interaction (DOI) information required for high-resolution imaging. The goal of positioning techniques is to reverse the photodetector signal’s pattern map to the coordinates of the incident photon energy position. By considering the DOI information, monolithic crystals offer good spatial, energy, and timing resolution along with high sensitivity. In this work, a supervised deep neural network was used for the approximation of DOI and to assess through Monte Carlo (MC) simulations the performance on a small-animal PET scanner consisting of ten 50 × 50 × 10 mm3 continuous Lutetium-Yttrium Oxyorthosilicate doped with Cerium (LYSO: Ce) crystals and 12 × 12 silicon photomultiplier (SiPM) arrays. The scintillation position was predicted by a multilayer perceptron neural network with 256 units and 4 layers whose inputs were the number of fired pixels on the SiPM plane and the total deposited energy. A GEANT4 MC code was used to generate training and test datasets by altering the photons’ incident position, energy, and direction, as well as readout of the photodetector output. The calculated spatial resolutions in the X-Y plane and along the Z-axis were 0.96 and 1.02 mm, respectively. Our results demonstrated that using a multilayer perceptron (MLP)-based positioning algorithm in the detector modules, constituting the PET scanner, enhances the spatial resolution by approximately 18% while the absolute sensitivity remains constant. The proposed algorithm proved its ability to predict the DOI for depth under 7 mm with an error below 8.7%.

Automation of data analysis in molecular cancer imaging and its potential impact on future clinical practice

Digitalization, especially the use of machine learning and computational intelligence, is considered to dramatically shape medical procedures in the near future. In the field of cancer diagnostics, radiomics, the extraction of multiple quantitative image features and their clustered analysis, is gaining increasing attention to obtain more detailed, reproducible, and meaningful information about the disease entity, its prognosis and the ideal therapeutic option. In this context, automation of diagnostic procedures can improve the entire pipeline, which comprises patient registration, planning and performing an imaging examination at the scanner, image reconstruction, image analysis, and feeding the diagnostic information from various sources into decision support systems. With a focus on cancer diagnostics, this review article reports and discusses how computer-assistance can be integrated into diagnostic procedures and which benefits and challenges arise from it. Besides a strong view on classical imaging modalities like x-ray, CT, MRI, ultrasound, PET, SPECT and hybrid imaging devices thereof, it is outlined how imaging data can be combined with data deriving from patient anamnesis, clinical chemistry, pathology, and different omics. In this context, the article also discusses IT infrastructures that are required to realize this integration in the clinical routine. Although there are still many challenges to comprehensively implement automated and integrated data analysis in molecular cancer imaging, the authors conclude that we are entering a new era of medical diagnostics and precision medicine.

Emerging methods in radiology

Imaging modalities have developed rapidly in recent decades. In addition to improved resolution as well as whole-body and faster image acquisition, the possibilities of functional and molecular examination of tissue pathophysiology have had a decisive influence on imaging diagnostics and provided ground-breaking knowledge. Many promising approaches are currently being pursued to increase the application area of devices and contrast media and to improve their sensitivity and quantitative informative value. These are complemented by new methods of data processing, multiparametric data analysis, and integrated diagnostics. The aim of this article is to provide an overview of technological innovations that will enrich clinical imaging in the future, and to highlight the resultant diagnostic options. These relate to the established imaging methods such as CT, MRI, ultrasound, PET, and SPECT but also to new methods such as magnetic particle imaging (MPI), optical imaging, and photoacoustics. In addition, approaches to radiomic image evaluation are explained and the chances and difficulties for their broad clinical introduction are discussed. The potential of imaging to describe pathophysiological relationships in ever increasing detail, both at whole-body and tissue level, can in future be used to better understand the mechanistic effect of drugs, to preselect patients to therapies, and to improve monitoring of therapy success. Consequently, the use of interdisciplinary integrated diagnostics will greatly change and enrich the profession of radiologists.

Machine Learning in PET: from Photon Detection to Quantitative Image Reconstruction

Machine learning has found unique applications in nuclear medicine from photon detection to quantitative image reconstruction. While there have been impressive strides in detector development for time-of-flight positron emission tomography, most detectors still make use of simple signal processing methods to extract the time and position information from the detector signals. Now with the availability of fast waveform digitizers, machine learning techniques have been applied to estimate the position and arrival time of high-energy photons. In quantitative image reconstruction, machine learning has been used to estimate various corrections factors, including scattered events and attenuation images, as well as to reduce statistical noise in reconstructed images. Here machine learning either provides a faster alternative to an existing time-consuming computation, such as in the case of scatter estimation, or creates a data-driven approach to map an implicitly defined function, such as in the case of estimating the attenuation map for PET/MR scans. In this article, we will review the abovementioned applications of machine learning in nuclear medicine.

(Hybrid) SPECT and PET Technologies

SPECT and PET are nuclear tomographic imaging modalities that visualize functional information based on the accumulation of radioactive tracer molecules. However, SPECT and PET lack anatomical information, which has motivated their combination with an anatomical imaging modality such as CT or MRI. This chapter begins with an overview over the fundamental physics of SPECT and PET followed by a presentation of the respective detector technologies, including detection requirements, principles and different detector concepts. The reader is subsequently provided with an introduction into hybrid imaging concepts, before a dedicated section presents the challenges that arise when hybridizing SPECT or PET with MRI, namely, mutual distortions of the different electromagnetic fields in MRI on the nuclear imaging system and vice versa. The chapter closes with an overview about current hybrid imaging systems of both clinical and preclinical kind. Finally, future developments in hybrid SPECT and PET technology are discussed.

Sensors for Positron Emission Tomography Applications

Positron emission tomography (PET) imaging is an essential tool in clinical applications for the diagnosis of diseases due to its ability to acquire functional images to help differentiate between metabolic and biological activities at the molecular level. One key limiting factor in the development of efficient and accurate PET systems is the sensor technology in the PET detector. There are generally four types of sensor technologies employed: photomultiplier tubes (PMTs), avalanche photodiodes (APDs), silicon photomultipliers (SiPMs), and cadmium zinc telluride (CZT) detectors. PMTs were widely used for PET applications in the early days due to their excellent performance metrics of high gain, low noise, and fast timing. However, the fragility and bulkiness of the PMT glass tubes, high operating voltage, and sensitivity to magnetic fields ultimately limit this technology for future cost-effective and multi-modal systems. As a result, solid-state photodetectors like the APD, SiPM, and CZT detectors, and their applications for PET systems, have attracted lots of research interest, especially owing to the continual advancements in the semiconductor fabrication process. In this review, we study and discuss the operating principles, key performance parameters, and PET applications for each type of sensor technology with an emphasis on SiPM and CZT detectors—the two most promising types of sensors for future PET systems. We also present the sensor technologies used in commercially available state-of-the-art PET systems. Finally, the strengths and weaknesses of these four types of sensors are compared and the research challenges of SiPM and CZT detectors are discussed and summarized.