Joint region-of-interest activity and alignment estimation in emission tomography

Previously, methods have been proposed for the direct reconstruction of the activity in a set of regions-of-interest (ROI), given the raw emission data and the locations of these ROIs in image space. These methods work well if the ROIs are well aligned, but artifacts may result if the alignment is poor. In this contribution, we propose a maximum likelihood reconstruction algorithm that estimates ROI-values where ROIs are available, and that applies standard voxel-based ML-reconstruction elsewhere. In addition, it jointly estimates the alignment of the ROIs to the image being reconstructed. The activity is updated with the standard MLEM update, the alignment with a least squares registration algorithm. A preliminary evaluation with simple 2D simulations and two microPET rat brain scans yielded promising results.