Design of an Architecture for Real-Time 3D PET Imaging

A major obstacle to the 3D approach in positron emission tomography is the time required to compute the reconstruction. The bulk of computations for no matter which algorithm is chosen is back-projection. This paper describes a possible architecture capable of supporting the scanner data rate while performing back-projection on an event-by-event basis. The architecture matches an event-by-event algorithm that allows parallelism and pipelining to be exploited in order to reduce the computational burden; it appears extremely suitable for back-projection filtering reconstruction algorithms.