A method of overlap correction for fully 3D OSEM reconstruction of PET data

In whole body PET studies, acquisitions are typically overlapped axially to improve the statistics of the slices at the edge of the field of view. This is particularly important in 3D acquisitions where the slice sensitivity profile varies more than in 2D acquisitions. 3D PET data in the overlapped region may be combined after reconstruction, by a weighted average of the corresponding images from the two frames, or the data may be combined after Fourier Rebinning is used to convert it to a stack of 2D transaxial sinograms. Direct combination of the 3D data on a frame by frame basis is difficult because many lines of response through the overlapped region also pass through the non-overlapped regions of image space. This paper describes a memory efficient method for combining overlapped 3D projection plane data in the loop of a fully 3D OSEM algorithm. In this method, a subset of 3D projection plane data is read from each frame and used to update the appropriate region of the ratio and correction images required in the iterative algorithm. The full 3D ratio and correction image volumes are then built for all frames for each subset. With each subset, the ratio and correction image volumes are used to create a new image estimate. This algorithm is compared to post reconstruction overlap correction using 3D OSEM on separate frames.