Performance improvement of event-based motion correction for PET using a PC cluster

Head motion during PET scanning produces significant artifact or spatial resolution loss on the reconstructed image. Event-based motion correction (EBMC) technique has been developed to correct head movement during the scan incorporated with list mode acquisition of PET and an optical tracking system. In EBMC technique, each line-of-response (LOR) in the list-mode data was reoriented due to the motion data by the optical tracking system. Although EBMC technique has potential to correct head movement during PET acquisition, large size of list mode data set hampers the capability of on-line processing for the correction. In order to improve in the speed of computing time for EBMC, we implemented EBMC on a Beowulf PC cluster consisting of 7 PCs (24 GHz Xeon for a master node and 1.4 GHz PentiumIII for slave nodes) connecting each other through Gbit Ethernet. MPI (Message Passing Interface) protocol was utilized for parallelizing the task of EBMC. The performance of the PC cluster was evaluated using list-mode data and head motion data acquired by ECAT EXACT HR+ (CTI/Siemens) PET scanner and POLARIS (Northern Digital) optical tracking system. The six list-mode data sets (file sizes are from 161 to 253 Mbytes) were corrected for motion by EBMC technique on a single PC and the PC cluster. The PC cluster was 5.2 times faster than the single PC to perform the motion correction. The PC cluster remarkably improves the performance of EBMC with low cost.