Brain Motion Correction for PET studies
Abstract:
In brain PET images, motion tends to blur the final images. As PET system technologies advance, this can be critical when using a high resolution PET system. As multiple motion events can occur in a short period of time, traditional methods perform poorly with PET short frame data as the images are typically noisy due to lack of statistics in the raw data. This paper introduces the Summing Tree Structure method that iteratively corrects for motion across motion frames by using 2D image projections to correct a 3D floating image to a 3D target image. In this approach, both the 3D floating and target images are initially projected onto the z axis, creating 2D projections. A spatial registration is calculated between the projections, and then applied to the 3D floating volume. This process is then repeated by projecting the registered 3D floating volume onto the y axis and registering to the 2D target projection, producing a new registered 3D volume; and then finally repeated again on the x axis. The method iterates until the floating image is sufficiently registered, and then it is summed to the target image. As there might be multiple motions detected within a brain PET study, the new summed image will become a floating image for a different motion tree node to register it to the node’s target image. This process can continue until all motion frames are registered to the final root tree target image.