Inverse Rendering for High Resolution PET Image Reconstruction with PSF and TOF Modeling

Incorporating point spread function (PSF) and time-of-flight (TOF) modeling into PET image reconstruction accounts for the physical nature of photon transport from object space to measurement space. In this study, we integrate PSF modeling within the inverse rendering-based reconstruction platform and introduce a modified digital differential analyzer (DDA) to accurately model TOF effects. The evaluation was conducted using Data Spectrum hotspot phantom and clinical tau PET data. In our proposed inverse rendering platform, forward projection is implemented through Monte Carlo estimation, where multiple points are randomly sampled within the crystals to define sub-lines of response (sub-LORs). For PSF modeling, Gaussian noise with zero mean is added to the sample positions to simulate spatial uncertainty in photon detection. Moreover, a modified DDA algorithm that weights voxel intensities according to the Gaussian kernel centered at TOF estimated position is used to incorporate TOF information. The phantom and clinical images were reconstructed using both CASToR and inverse rendering platform. The line profile of 1 mm diameter hotspots indicated our platform achieved a 33% higher average peak-to-valley ratio (PVR) compared to CASToR. Clinical results further demonstrated reduced partial volume effects and higher contrast in the regions of interest. These improvements support the inverse rendering platform's potential for accurate tau PET imaging and improved Braak staging in Alzheimer's disease assessment.