Optimization of temporal basis functions in dynamic PET imaging

In this paper we introduce and evaluate a completely data-driven method to optimize both the linear expansion coefficients and the used temporal basis functions in dynamic PET reconstruction from list-mode data. sent the first results of our method using simulated 2D PET data. The time activity curves are modeled as a conic combination of B-splines. The B-splines are optimized with respect to their knot locations. We have combined the newly introduced method with our previously developed dynamic MLEM algorithm in order to further improve the likelihood. The results show that even one iteration of our newly developed algorithm can drastically increase the likelihood and the resulting images better represent the underlying TACs.