Data-driven patient motion estimation for PET brain imaging

205 Objectives With the advent of the new F-18 labeled Alzheimer-imaging agents quantitative PET brain imaging has increased in clinical importance. One factor which impacts the accuracy of such imaging is head-motion during imaging. We are therefore investigating motion correction for PET brain imaging based on solely the acquired PET data (data-driven). Herein we investigate subdividing the clinical list-mode PET acquisitions into increasingly smaller time-bins to determine the impact from using different duration of time-buns and the estimation accuracy as motion becomes more frequent. Methods Clinical list-mode PET studies in which there was no motion beyond small gradual drifts were subdivided into increasingly smaller time-bins The bins were independently reconstructed and small rigid-body 6-degree-of-freedom (6-DOF) known movements were applied to selected bins to simulate known patient motion. The reconstructed time bins were pre-processed to reduce noise and to blur out the activity variation in the interior, so that the registration primarily used the outer shape of activity distribution. Motion between the first bin and subsequent bins was estimated by the ITK multi-resolution image registration algorithm [1]. Results For 10 s bins, translation was estimated with sub-pixel accuracy ( Conclusions Time-bins as short as 5 seconds may be used for data-driven motion estimation of the brain in PET. Research Support NIH R01-EB001457 and Philips Healthcare