ML reconstruction from dynamic list-mode PET data using temporal splines

We implemented and evaluated a maximum likelihood-optimality condition iteration algorithm (ML-OCI) to reconstruct dynamic PET data. The time activity curves (TACs) were reconstructed on a spatially segmented image. The segmented image paradigm effectively cancels out spatial reconstruction issues allowing a time domain evaluation of our method. The TACs were represented on a B-spline basis. We investigated different parameters of this basis such as order, number of basis functions and knot placing in a reconstruction task, using simulated dynamic list-mode data. We found that a higher density of basis functions allows the algorithm to follow faster changes in the TAC, however the TACs become noisier. Therefore an adaptive knot placing strategy is developed and evaluated. It allowed a more accurate reconstruction while preserving the same noise-level.