Patlak reconstruction from dynamic PET based on diffusion models: evaluations using total-body dynamic datasets

Dynamic PET can quantitatively recover physiology-related parameters. It is widely used in research and is being adopted in clinics. For dynamic PET, kinetic modeling is needed to reconstruct parametric images based on a specific kinetic model. The parametric images reconstructed always suffer from low image quality, mainly due to the ill-posedness of kinetic-model fitting and limited counts due to non-continuous data acquisition for each bed position. In this work, we proposed a diffusion model-based kinetic-model fitting framework, with Patlak model as one example. For this proposed framework, the score function was pre-trained based on static total-body PET images and utilized as the prior for Patlak slope and intercept images, considering the patch-wise similarity among them. During inference, the Patlak kinetic model was employed as a data-consistency constraint to generate the final Patlak images. Preliminary results based on total-body dynamic datasets demonstrated the feasibility and promising performance of the proposed framework.