Inter-Frame Co-Registration of Dynamically Acquired Fluoro-Deoxyglucose Positron Emission Tomography Human Brain Data

When acquiring positron emission tomography (PET) brain scans dynamically from a patient, it would be helpful to be able to minimize confounding effects of his/her head movement. In this study, we attempted to coregister sequential frames in the dynamic fluorodeoxyglucose (FDG) PET image sequence, specifically addressing the challenge associated with the relatively low radiotracer uptake over the cortical areas in the early short frames. PET data were acquired dynamically over sixty minutes in three healthy volunteers (frame durations ranged from 2 seconds to 5 minutes). We realigned each of the early short frames to its immediate predecessor using normalized mutual information (NMI). Afterward, the same procedure was repeated to realign the summation of the early frames, as well as each of the later frames with durations of 5 minutes and sufficient cortical tracer uptake. To validate our proposed procedure, we artificially moved frames using known linear rotations and translocations and compared the 6 NMI estimated rigid-body transformation parameters to the actual misalignment parameters and compared image-derived carotid artery FDG input function in the realigned and original images. While counting statistics in the first 3 frames (22 seconds after FDG administration) precluded us from accurately addressing misalignments, the NMI reliably recovered translocations as great as 7 mm and rotations as great as 8 degrees for all other frames. Moreover, misalignment errors in the constructed input function from the carotid artery region were able to be corrected. Our NMI-based strategy for the realignment of sequential frames appears to provide a helpful way to determine the severity of the head movement during a PET scan and correct for the effects. Additional studies are needed to address additional confounds associated with the misalignment between transmission and emission scans.