Joint Reconstruction of Activity and Attenuation for PET Imaging with Diffusion Prior

Attenuation correction is essential for accurate quantification in PET imaging. The popular Maximum Likelihood reconstruction of Activity and Attenuation (MLAA) algorithm estimates the attenuation maps and PET activity images jointly from PET sinograms but often suffers from slow convergence and high noise levels. In this work, we developed a diffusion model-based joint reconstruction algorithm that incorporates a diffusion model into the joint reconstruction of attenuation and activity, termed DEMAA. DEMAA requires only a dif-fusion model pre-trained on the widely available CT images, eliminating the need for paired datasets or tracer-specific training. Our preliminary results show that DEMAA significantly accelerated convergence in terms of Root Mean Squared Error (RMSE) and demonstrated potential for jointly estimating attenuation map and activity from non-TOF PET sinogram.