Using PCA to detect head motion from PET list mode data

Abstract

PET can be used for studying the brain. However, structures in the brain are small and results can be affected by patient motion, especially during dynamic PET scans. Many researchers have studied correction of head movement using data from an external camera. Here, we report on a fast technique to detect when the movement happens from the PET data only. The method uses PCA on (spatially coarse) dynamic sinogram data. It allows the detection of movements in arbitrary direction, as long as the movement is fast compared to biological kinetics. We use this information as input for the Multiple Acquisition Frame (MAF) technique to split the list mode data in dynamic frames, ignoring periods of fast movement, followed by image reconstruction and rigid registration. We compare this technique with using dynamic data with fixed time frames of 30secs or 300secs. We show that for a given number of time frames, MAF-PCA can give better results than using fixed frame duration.