Positron Emission Tomography (PET) is a molecular imaging technique that creates an image of radiopharmaceutical distribution using acquired sinogram data. Accurate PET reconstruction requires a balance between data fidelity and regularization. The choice of penalty type and the adjustment of the regularization parameters have a critical effect on image quality, including noise suppression, edge preservation, and contrast recovery. However, manually determining optimal regularization strength is challenging due to their dependence on data properties and clinical needs. It often requires multiple reconstructions and considerable time to achieve satisfying results. To address this issue, we propose SINORES, a supervised deep learning approach to predict the optimal regularization parameters for the modified Block Sequential Regularized Expectation Maximization (BSREM) algorithm. The prediction is based on the raw sinogram data and the scaling coefficients that encode acquisition-related properties. By learning from a synthetic dataset of 2D sinograms and scaling coefficients paired with their corresponding optimal parameters, SINORES rapidly identifies suitable parameter values and avoids the need for manual setting. This work presents a proof of concept that demonstrates the feasibility of the proposed framework in the context of 2D PET imaging. The proposed method achieves consistent parameter estimation across different phantom types and effectively determines suitable parameters for reconstructing real PET data, leading to improved reconstruction quality and reliability in practical settings.

Implicit neural representations (INRs) have recently emerged as a continuous and flexible approach for PET image reconstruction, mitigating artifacts produced by conventional algorithms. In this work, we introduce SPINR, an enhanced implicit neural representation framework for self-supervised PET reconstruction. SPINR features a refined coordinate normalization strategy that enables spatial encoding to depend on the field of view rather than fixed image coordinates, thereby improving robustness across diverse reconstruction settings. Additionally, we conduct a spectral and performance analysis of multiple periodic and non-periodic activation functions to better understand the frequency-encoding behavior of INR architectures in PET imaging. Experimental results show that SPINR delivers good performance compared to both traditional penalized-likelihood reconstructions and deep image prior (DIP)-based approaches, providing improved contrast recovery, reduced bias, and enhanced overall image quality. Collectively, these findings establish SPINR as a dataefficient and robust framework for PET image reconstruction.

BACKGROUND:Scatter correction is essential for quantitative and accurate time-of-flight (TOF) PET imaging. It is implemented by an accurate scatter estimation algorithm, to calculate the statistical distribution of scattered photons among the measured coincidences. However, to our knowledge, scatter estimation algorithms that account for TOF and that are compatible with custom geometries are not available in open-source reconstruction libraries, such as CASToR and STIR. To this end, we have developed an open-source implementation of the TOF-aware single-scatter-simulation (SSS) algorithm: openSSS. RESULTS:openSSS is validated on NEMA phantoms and patient data, for three PET geometries, compared to Monte-Carlo simulations and two proprietary vendor-specific reconstruction platforms. The reconstructed images have similar contrast recovery and background variability, deviating by up to 3.7%-point on contrast recovery and 1.8 on background variability and looking visually similar. CONCLUSION:We have developed and validated an open-source scatter estimation library to complement reconstruction frameworks. By enabling vendor-independent clinical-grade reconstructions on custom scanner geometries, openSSS represents a crucial step in transparent research on quantitative PET and novel PET scanner designs.

Implicit neural representations (INRs) have demonstrated strong capabilities in various medical imaging tasks, such as denoising, registration, and segmentation, by representing images as continuous functions, allowing complex details to be captured. For image reconstruction problems, INRs can also reduce artifacts typically introduced by conventional reconstruction algorithms. However, to the best of our knowledge, INRs have not been studied in the context of PET reconstruction. In this paper, we propose an unsupervised PET image reconstruction method based on the implicit SIREN neural network architecture using sinusoidal activation functions. Our method incorporates a forward projection model and a loss function adapted to perform PET image reconstruction directly from sinograms, without the need for large training datasets. The performance of the proposed approach was compared with that of conventional penalized likelihood methods and deep image prior (DIP) based reconstruction using brain phantom data and realistically simulated sinograms. The results show that the INR-based approach can reconstruct high-quality images with a simpler, more efficient model, offering improvements in PET image reconstruction, particularly in terms of contrast, activity recovery, and relative bias.

We read with great interest the paper by Lim et al (2018 Phys. Med. Biol. 63 035042) on bias reduction in Y-90 PET imaging. In particular, they proposed a new formulation of the tomographic reconstruction problem that enforces non-negativity in projection space as opposed to image space. We comment on the algorithm they derived from this formulation and bring some clarifications on the constraint that this algorithm respects.

Positron Emission Tomography (PET) is a medical imaging modality relying on numerical methods that integrate the statistical properties of the measurements and prior assumptions about the images. In order to maximize the computed image quality, PET reconstruction algorithms require the setting of hyperparameters that balance data fidelity with regularization. However, their optimal tuning depends on the statistical properties of the raw data and on the clinical objectives. To address this issue, we propose a supervised deep learning strategy based on a residual neural network that takes the raw measured data (sinogram) as input and automatically predicts the optimal value of the regularization parameter of the modified block Sequential Regularized Expectation Maximization (BSREM) algorithm. The proposed strategy is trained on a synthetic dataset consisting of 2D sinograms and their corresponding optimal regularization parameters. Our results demonstrate the feasibility of the approach leading to improved image reconstruction compared to classical manual tuning methods.

Multiplexed PET imaging offers the possibility to capture the diverse physiological and pathological characteristics of tissues by employing multiple radiotracers simultaneously. This study introduces a method for direct reconstruction and separation of images from mixed sinograms. The approach aims to be generic and unsupervised in order to be suitable for future clinical applications. The proposed method is compared to the indirect method previously published and found to have better performances closer to those obtained with separate singletracer acquisitions

Deep Image Prior (DIP) is an unsupervised way of denoising images with a neural network, trained only using the noisy image. DIP is currently used in Positron Emission Tomography (PET), embedded in the PET reconstruction or directly for PET reconstruction. The original authors used a U-Net architecture with skip connections (SC) and random noise as input. A slightly modified U-Net is usually used in PET, especially with an anatomical image replacing the random input. In this work, we studied the impact of the choice of the input image with the presence or absence of SC. We considered the DIP embedded as a constraint into the PET reconstruction using the DIP Nested ADMM (DNA) and the DIPrecon algorithms. Results demonstrated that anatomical mismatches or noise coming from the chosen input may appear in output images using a network with too many SC, whereas few SC led to distorted regions. An intermediate setting where the anatomical image is used as an input only for the initialization of the network, and then replaced by random noise within the iterations, achieved the best results without mismatch artifacts, and appeared less sensitive with respect to the number of SC.

This contribution addresses the problem of image reconstruction of radioactivity distribution for which the available information arises from several classes of data, each associated with a specific combination of detections. We introduce a theoretical framework to measure the amount of information brought by each class and we develop an iterative algorithm dedicated to multi-class reconstruction based on maximum likelihood.We apply our approach to the XEMIS2 camera, a preclinical prototype of a Compton telescope dedicated to 3-photon PET imaging for which four distinct classes of partial detections coexist with the full detection class.Based on Monte Carlo simulations, we present the first elements of our model.

As the capabilities of computer hardware and software progresses, the importance of standardized data is becoming more relevant. While DICOM is widely used as a standard for PET images, no agreed format exists for PET raw data. In addition, all current vendor formats differ, to accommodate different architectures and processing strategies.An international working group comprised of PET imaging experts from academia and industry was formed in 2022 to propose a new standard format. Efforts include establishing the key informational elements to include in the standard, data container formats, and integration initiatives. An important feature of the proposal is the use of a meta-language (called Yardl) for defining data structure and protocols for accessing, transferring, and storing data, allowing logical separation of data elements and container formats. All portions of the format, tools for accessing the data in the standardized format, and example data will be made open source and publicly available.In this work, we will provide a status report of the progress of the working group. The effort is ongoing, and we welcome interest, feedback, and support from the greater community. We hope that this standard will facilitate a new paradigm for PET innovation, including new opportunities for inter-scanner and inter-vendor harmonization and AI applications, and aide in the advancement of novel PET applications and analysis tools.

Using different tracers in positron emission tomography (PET) imaging can bring complementary information on tumor heterogeneities. Ideally, PET images of different tracers should be acquired simultaneously to avoid the bias induced by movement and physiological changes between sequential acquisitions. Previous studies have demonstrated the feasibility of recovering separated PET signals or parameters of two or more tracers injected (quasi-)simultaneously in a single acquisition. In this study, a generic framework in the context of dual-tracer PET acquisition is proposed where no strong kinetic assumptions nor specific tuning of parameters are required. The performances of the framework were assessed through simulations involving the combination of [18F]FCH and [18F]FDG injections, two protocols (90 and 60 min acquisition durations) and various activity ratios between the two injections. Preclinical experiments with the same radiotracers were also conducted. Results demonstrate the ability of the method both to extract separated arterial input functions (AIF) from noisy image-derived input function and to separate the dynamic signals and further estimate kinetic parameters. The compromise between bias and variance associated with the estimation of net influx rateKishows that it is preferable to use the second injected radiotracer with twice the activity of the first for both 90 min [18F]FCH+[18F]FDG and 60 min [18F]FDG+[18F]FCH protocols. In these optimal settings, the weighted mean-squared-error of the estimated AIF was always less than 7%. TheKibias was similar to the one of single-tracer acquisitions; below 5%. Compared to single-tracer results, the variance ofKiwas twice more for 90 min dual-tracer scenario and four times more for the 60 min scenario. The generic design of the method makes it easy to use for other pairs of radiotracers and even for more than two tracers. The absence of strong kinetic assumptions and tuning parameters makes it suitable for a possible use in clinical routine.

Radioembolization with 90 Y-microspheres is used as a treatment for non-resectable liver cancer. 90 Y is mainly a β- emitter but a few β+ particles are also emitted. It enables to quantify the amount of radioactivity in the body using PET imaging, which could be especially useful for dosimetry purpose. Yet, the reconstructed images are very noisy due to the limited amount of collected data with 90 Y, and the usual reconstruction algorithms have positive bias in regions with low activity. In this context, we propose to combine two complementary approaches recently published, both using the Alternating Direction Method of Multipliers (ADMM) algorithm, within a nested ADMM. The first one allows for negative values in the image by enforcing the non-negativity in the projection space only, hence reducing the bias. The second one intends to lower the noise in the image by adding the constraint that the reconstructed image is the output of a Deep Image Prior (DIP) network.

Modélisation et reconstruction de cartes paramétriques corps-entier en imagerie pharmacologique TEP-IRM La tomographie par émission de positons (TEP) est fréquemment utilisée pour des applications cliniques, avec une majorité des pratiques reposant sur des mesures qualitatives et semi-quantitatives. Mais l'imagerie TEP a la capacité de fournir des informations fonctionnelles entièrement quantitatives sur les processus sous-jacents explorés, grâce à l'imagerie dynamique et à la modélisation cinétique. Ces informations quantitatives peuvent être utilisées comme biomarqueurs pour des applications cliniques, en particulier pour la médecine de précision. Des protocoles avec des positions du lit multiples, dédiés à l'imagerie dynamique du corps entier (DWB), ont été développés afin étendre le champ de vue effectif, au prix de restrictions dans le nombre d'acquisitions et la fréquence d'échantillonnage. L'objectif de cette thèse est d'améliorer la qualité de l'imagerie paramétrique du corps entier pour les applications d'imagerie DWB sur un système hybride TEP-IRM. Dans notre première contribution, nous avons présenté le développement d'un protocole entièrement automatisé pour l'imagerie DWB sur un système TEP-IRM clinique, qui a permis de réduire les délais d'acquisition, ce qui se traduit par une augmentation du nombre d'acquisitions et de la fréquence d'échantillonnage. Le recours à l'automatisation complète a permis d'optimiser la planification des positions individuelles des lits, en utilisant au mieux le champ de vue effectif. Pour la deuxième contribution, nous avons développé des algorithmes de reconstruction dynamique dans un logiciel ouvert, et évalué les avantages offerts par l'utilisation de divers modèles cinétiques dans la reconstruction de données TEP dynamiques simulées et réelles. Nos résultats sont en accord avec les conclusions d’études antérieures sur l'utilisation de la reconstruction dynamique. Dans notre cas particulier de l'imagerie DWB, la reconstruction dynamique a montré des propriétés favorables pour l'exactitude et la précision des images paramétriques du corps entier, tout en fournissant des images dont le bruit est comparable à celui des protocoles dynamiques standards à position de lit unique, reconstruits avec des techniques ordinaires. Dans notre troisième contribution, nous présentons une extension des fonctionnalités développées précédemment : la reconstruction dynamique simultanée de toutes les données multi-lits. Cette méthodologie permet l'utilisation synchrone de toutes les données d'acquisition DWB dans une seule boucle de reconstruction. La méthode a été appliquée à une étude pharmacocinétique DWB réalisée sur un système TEP-IRM clinique. Une comparaison a été faite avec des reconstructions statiques standards suivies d'une modélisation cinétique post reconstruction. Les résultats obtenus avec les deux méthodes étaient en bon accord, sans introduction de biais sur les métriques évaluées. En outre, l'utilisation de la reconstruction dynamique a entraîné une réduction notable du bruit dans les images d’émission et paramétriques. En outre, une méthode de détection et de correction des erreurs de modélisation utilisant la modélisation résiduelle adaptative a été appliquée et évaluée. Elle a montré des résultats prometteurs pour la réduction des erreurs de modélisation et leur propagation, tout en permettant la généricité dans l'utilisation des algorithmes de reconstruction dynamique. Nos résultats ont montré que la reconstruction dynamique est nécessaire en imagerie paramétrique corps-entier pour obtenir une quantification précise et stable. De nombreuses méthodes ont été proposées dans ce projet afin d’optimiser le processus de reconstruction TEP pour l'imagerie DWB, en utilisant au mieux les données dynamiques acquises sur plusieurs positions de lit. Pour généraliser son utilisation, certaines améliorations méthodologiques doivent encore être apportées pour garantir une imagerie paramétrique fiable et sans artefact, notamment en ce qui concerne les mouvements du patient.

The popularity of yttrium-90 ( 90 Y) PET is growing. However, due to the very low branching ratio of 90 Y ( $3.2\times 10^{-5}$ ), images reconstructed are characterized by a high noise level and positive bias in low-activity regions. To overcome this problem, some algorithms use penalized reconstructions, and others allow for negative values in the image. Recently, a post-processing method has also been proposed that removes the induced negative values while maintaining bias reduction. The work presented in this article aims to evaluate and compare these methods to guide the reader in the choice of the best-suited reconstruction algorithm and associated parameters. First, several algorithms were tested using experimental phantom data. Pareto fronts and Pareto sets from the multiobjective optimization formalism were used for the comparison. Next, a dosimetric study (using phantom and patient data) was conducted to assess the final impact of these algorithms. The lowest biases were reached by unconstrained algorithms. When compared to penalized algorithms, the latter allowed for noise reduction at a fixed level of bias, with the best results obtained using a penalty based on relative differences. A good compromise between noise and bias could be reached by combining penalty, early stopping of iterative algorithms, unconstrained algorithms, and post-processing.

Dynamic whole body (DWB) PET acquisition protocols enable the use of whole body parametric imaging for clinical applications. In FDG imaging, accurate parametric images of Patlak K i can be complementary to regular standardised uptake value images and improve on current applications or enable new ones. In this study we consider DWB protocols implemented on clinical scanners with a limited axial field of view with the use of multiple whole body sweeps. These protocols result in temporal gaps in the dynamic data which produce noisier and potentially more biased parametric images, compared to single bed (SB) dynamic protocols. Dynamic reconstruction using the Patlak model has been previously proposed to overcome these limits and shown improved DWB parametric images of K i . In this work, we propose and make use of a spectral analysis based model for dynamic reconstruction and parametric imaging of Patlak K i . Both dynamic reconstruction methods were evaluated for DWB FDG protocols and compared against 3D reconstruction based parametric imaging from SB dynamic protocols. This work was conducted on simulated data and results were tested against real FDG dynamic data. We showed that dynamic reconstruction can achieve levels of parametric image noise and bias comparable to 3D reconstruction in SB dynamic studies, with the spectral model offering additional flexibility and further reduction of image noise. Comparisons were also made between step and shoot and continuous bed motion (CBM) protocols, which showed that CBM can achieve lower parametric image noise due to reduced acquisition temporal gaps. Finally, our results showed that dynamic reconstruction improved VOI parametric mean estimates but did not result to fully converged values before resulting in undesirable levels of noise. Additional regularisation methods need to be considered for DWB protocols to ensure both accurate quantification and acceptable noise levels for clinical applications.

The uncertainty of reconstructed PET images remains difficult to assess and to interpret for the use in diagnostic and quantification tasks. Here we provide (1) an easy-to-use methodology for uncertainty assessment for almost any Bayesian model in PET reconstruction from single datasets and (2) a detailed analysis and interpretation of produced posterior image distributions. We apply a recent posterior bootstrap framework to the PET image reconstruction inverse problem and obtain simple parallelizable algorithms based on random weights and on existing maximum a posteriori (MAP) (posterior maximum) optimization-based algorithms. Posterior distributions are produced, analyzed and interpreted for several common Bayesian models. Their relationship with the distribution of the MAP image estimate over multiple dataset realizations is exposed. The coverage properties of posterior distributions are validated. More insight is obtained for the interpretation of posterior distributions in order to open the way for including uncertainty information into diagnostic and quantification tasks.

In PET imaging, the use of different tracers may provide complementary information on tumor heterogeneities. A single PET acquisition with dual-tracer injection prevents bias between sequential acquisitions. As already shown in the literature, it is possible to separate two PET signals from equal half-life isotope tracers based on their pharmacokinetics. With the goal of building a generic framework for reconstructing separated images of each tracer from dual-tracer PET acquisitions, we propose to take benefits from a spectral model without any assumptions about kinetics. Using 1D simulations of FLT+FDG, we evaluated the ability of the model to separate and extract parameters of interest for each tracer with respect to the delay between injections.