Event-by-event motion compensation in 3D PET

Motion compensation in brain PET using a multiple acquisition frame (MAF) method in conjunction with an optical motion tracking system has been previously described. The MAF method is applicable to multi-frame PET data acquired dynamically, and also to list mode data that have been sorted into multiple frames. The MAF method compensates very effectively for frame-to-frame motion, but cannot compensate for intra-frame motion, i.e. motion that occurs during the acquisition of a frame. We have investigated the use of list mode acquisition and LOR rebinning (in which individual list mode events are spatially transformed to compensate for changes in head position) as a way of overcoming the limited temporal discrimination of the MAF method. The feasibility of LOR rebinning was assessed in a list mode scan of the Hoffman brain phantom in which multiple 6 degree-of-freedom (d.f.) movements were applied to the phantom during acquisition. The LOR-rebinned data were stored in a 3D sinogram, rebinned to 2D with Fourier rebinning (FORE) and reconstructed with filtered backprojection (FBP). The resulting images exhibited a marked reduction in distortion, but appeared noisier and had poorer contrast than motion-corrected images produced with the MAF method. This was attributed to the loss of events during LOR rebinning (24% of the total), and spatial approximations in associating transformed LORs with sinogram elements. LOR rebinning combined with list mode ML-EM reconstruction was explored as an alternative method because it does not require transformed events to be stored in sinogram format, and thus avoids count losses and spatial approximations. Results from an initial implementation of a list mode ML-EM reconstruction algorithm are presented.