Organ concentration quantification for small animal PET images by registration with a statistical mouse atlas

This study focuses on quantification of probe concentration in major organs of small animal micro-PET images. In order to delineate organ ROIs, a statistical mouse atlas is registered to the micro-PET image. This statistical atlas is trained from 22 organ-labeled micro-CT images using Principle Component Analysis (PCA). By tuning the shape-controlling parameters of the atlas, we are able to adapt the atlas anatomy to fit the individual subject of the micro-PET image. To reduce the spillover effect, voxel confidence values are computed and are used to calculate confidence-weighted mean concentration of each organ. Experiments were performed on both simulated images and in-vivo micro-PET emission images. Results showed that the registration-based yielded comparable accuracy to the quantification based-on ground truth organ regions, and the confidence-weighted averaging obtains more accurate results than direct averaging of organ concentrations.