Skip to Main Content
Dynamic (4-D) PET imaging is a routinely used technique applied in clinical studies, and is now also finding application for high resolution animal PET systems. The usual method of implementing the image reconstruction involves separating the acquired data out into lower count discrete time frames, resulting in increased image noise. This paper presents two further developments for one-pass 4-D list-mode expectation-maximization (EM) reconstruction, which has already demonstrated enhanced reconstruction quality for image sequences compared to conventional strategies, as well as significantly reducing reconstruction time. The technique, which is an extension of one-pass 3-D list-mode EM, filters a previously reconstructed frame, which is then used as a starting image for the following frame, thus including prior object information. A new approach to post-reconstruction smoothing is also proposed, which in some cases is able to reduce noise without loss of resolution. Using measured quad-HIDAC data from a moving point source and a specially designed liquid flow experiment, and a rat scan, the two approaches are compared to the uniform image initialization method in terms of spatial and temporal resolution, noise, and accuracy of time-activity curves. Preliminary results, using resolution recovery, indicate the potential of achieving ultra-high spatial resolution with good temporal resolution, arising from improving the use of otherwise wasted prior information.