부산대학교 도서관

Motion in PET studies degrades the image quality and introduces bias which is critical for high resolution scanners. There are many publications related to motion correction in PET. Most of these methods rely on external devices to track the motion and register it to the listmode data. This paper describes how to extract rigid-body head motion from only the listmode data using the Centroid Of Distribution (COD). In addition, it introduces the Adjacent-Means Clustering and Merging of Clusters method to allocate in time the motion occurrences and the displacement values of the rigid body. The output resulting from the COD is the (x,y,z) positioning, and it is used to generate an (n) number of clusters based on an initial positioning starting point and a delta displacement of each point ε. The ε value can be variable based on the number of clusters required. The method frames each cluster with a start and end point and provides it with an incremental unique ID. Once the initial clusters are identified, the mean value of each cluster’s displacement is calculated and reused as the initial starting point for the next iteration. The mean displacement points of adjacent clusters are compared and merged if their delta displacement is within the identified ε, and the merged cluster mean values are calculated. Once the algorithm passes through the merging of the clusters, the method reiterates using the new initial starting point to pass through the COD data. As the initial starting point changes, the number of clusters will also decrease as points in previous iteration clusters merge into the new adjacent clusters. After i iterations, the output will be the number of merged clusters and their mean position. The resulting clusters represent motionless frames with a start and end point in time that can be used for framing the listmode data.