MuPET (Multiphysics PET): Bridging Computational Mass Transport Physics and Medical Imaging for Synthetic Positron Emission Tomography Data Generation
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Purpose
Robust positron emission tomography (PET) research increasingly requires reliable ground-truth activity distributions, yet clinical data and physical phantoms cannot provide them fully. This work introduces MuPET (multiphysics PET), a computational framework that bridges mass-transport physics simulation with image domain noise modeling to generate realistic, synthetic PET images.
Methods
Tracer transport and radioactive decay within a custom phantom geometry were modeled via decay diffusion physics. The simulated activity concentrations were converted into PET equivalent count maps, followed by a pseudo-Poisson noise model calibrated to scanner-specific characteristics. MuPET generated images were compared with experimental PET acquisitions and a conventional synthetic lesion insertion pipeline. VOI (volume of interest) and ROI (region interest) data extraction was performed on all the data. The quantitative metrics included the recovery coefficient (RC), target contrast ratio (TCR), coefficient of variability (COV), and activity profile.
Results
The simulated effective activities matched the decay-corrected measurements within 0.4–1.2%, confirming high physics-model fidelity. The optimized noise model achieved COV values (3.9–4.0%) comparable to those of the acquisitions (3.8–3.9%). MuPET demonstrated strong agreement with the experimental profiles and quantitative metrics, with mean biases within 10% of the acquisition.
Conclusion
MuPET establishes a fast (approximately 15 min), physically grounded pipeline that combines transport modeling with PET image simulation, enabling generation of realistic datasets. This approach provides a practical foundation for future dynamic PET simulations.
