PET Rapid Image Reconstruction Challenge (PETRIC)

TL;DR

PETRIC introduces the first public benchmarking challenge for PET image reconstruction focused on minimizing runtime while achieving target image quality. It formalizes a MAP reconstruction problem with a smoothed RDP prior, provides a BSREM reference, and uses a shared open-source framework (SIRF/CIL) to ensure fair comparisons across heterogeneous phantom datasets. Four teams contributed nine gradient-based algorithms, revealing speed-accuracy trade-offs and the sensitivity of performance to preconditioning and step-size choices, while promoting reproducibility through open code and Docker images. The study highlights practical considerations for real-data benchmarking, data curation, and the potential to extend the framework to future challenges and deeper-learning approaches.

Abstract

Introduction: We describe the foundation of PETRIC, an image reconstruction challenge to minimise the computational runtime of related algorithms for Positron Emission Tomography (PET). Purpose: Although several similar challenges are well-established in the field of medical imaging, there have been no prior challenges for PET image reconstruction. Methods: Participants are provided with open-source software for implementation of their reconstruction algorithm(s). We define the objective function and reconstruct "gold standard" reference images, and provide metrics for quantifying algorithmic performance. We also received and curated phantom datasets (acquired with different scanners, radionuclides, and phantom types), which we further split into training and evaluation datasets. The automated computational framework of the challenge is released as open-source software. Results: Four teams with nine algorithms in total participated in the challenge. Their contributions made use of various tools from optimisation theory including preconditioning, stochastic gradients, and artificial intelligence. While most of the submitted approaches appear very similar in nature, their specific implementation lead to a range of algorithmic performance. Conclusion: As the first challenge for PET image reconstruction, PETRIC's solid foundations allow researchers to reuse its framework for evaluating new and existing image reconstruction methods on new or existing datasets. Variant versions of the challenge have and will continue to be launched in the future.

Figures (4)

• Figure 1: Reference images for two phantoms: on the left the low counts NEMA scanned with the Mediso clinical PET scanner, on the right the Hoffman phantom scanned with a Siemens Biograph Vision 600. The images shown are cropped to the location of the phantom.
• Figure 2: Mean rank of the algorithms across all datasets. The black bars show the standard error in the mean.
• Figure 3: Difference of solution obtained by all algorithms stopped when the winning algorithm reached the threshold and reference image, for the NEMA phantom low counts acquired with a Mediso scanner. The plots are presented in a diverging color map with white indicating a match, blue indicating underestimation and red overestimation with respect to the reference solution. The scale of the color map is 10% of the maximum value of the reference image and is the same for all plots. The TomoUnimib algorithm did only 3 iterations and it was diverging with a result that was still the zero image. The images shown are cropped to the location of the phantom.
• Figure 4: Difference of solution obtained by all algorithms stopped when the winning algorithm reached the threshold and reference image, for the Hoffman phantom scanned with the Siemens Biograph Vision600. The plots are presented in a diverging color map with white indicating a match, blue indicating underestimation and red overestimation with respect to the reference solution. The scale of the color map is 10% of the maximum value of the reference image and it is the same for all plots. Here all algorithms except MaGeZ's produced a smoother image than the reference, as indicated by the seemingly random noise visible in rows 2 and 3; however, there is some structural difference in the difference plots, especially between the low and high intensity regions. The images shown are cropped to the location of the phantom.