Incorporating frame-to-frame coupling in simultaneous reconstruction of dynamic image sequences in PET

In dynamic PET studies, cross-sectional images of individual frames are reconstructed separately. Based on the general observation that images of adjacent frames do not change abruptly, we have explored a method to reconstruct all images of a dynamic sequence simultaneously that can incorporate frame-to-frame coupling to improve signal-to-noise level of the reconstructed images. In this study, frame-to-frame coupling is introduced by including in the objective function a penalty term that is proportional to the square of the pixel value difference between adjacent frames. The approach together with a total variation prior (for inplane regularization) in maximum likelihood reconstruction was found to achieve a significant reduction in image noise level, especially for images of early frames of short scan duration. The effectiveness of the method for signal-to-noise improvement is demonstrated with computer simulated data and with a real brain FDG study.