Comparison between an image- and a sinogram-based correction algorithm for partial volume effect in 3D PET imaging

Two fully 3D partial volume correction (PVC) techniques in PET imaging are compared. They follow the region based method proposed in 2D by O. Rousset et al. (1998). They assume that the object being imaged consists of anatomical domains with homogeneous true activity and that the voxel intensity in the PET image is the sum of the true activity in each domain weighted by its regional spread function (RSF). The two implementations that we compare differ in the way the RSFs are obtained: (1) a 3D extension of the original work of Rousset, that is based on an analytical simulator, and (2) a convolution of the anatomical tissue domains, in the image space, with the 3D PET system PSF. We used a Monte Carlo simulated cerebral dynamic study to assess the performance of both PVC implementations in the recovery of the time activity curves for the striata. The two methods allow the recovery of the true time activity curves with RMS errors of about 4%. The advantage of the second approach is its simplicity and rapidity that would enable fully 3D PVC in a clinical context, for protocols dedicated to compartmental analysis that require a few accurate ROI time activity curves.