Atlas-guided automated analysis of small-animal PET studies

This work aims to develop a methodology for automated atlas-guided analysis of small animal PET data through deformable registration to an anthropomorphic mouse atlas. A non-rigid registration technique is used to put into correspondence relevant anatomic regions of rodents to the predefined atlas. The technique consists in registering CT images from actual mouse PET/CT data to corresponding CT images of the Digimouse atlas, thus providing a pre-segmented anatomical model consisting of 21 anatomical regions suitable for automated analysis. Image registration is performed using the Elastix package, which is a modular toolbox based on the ITK library allowing the implementation of various image registration algorithms. Once the optimal parameters were derived, these were applied to all data sets. The accuracy of image registration was assessed by segmenting mice CT images into 7 regions: brain, lungs, heart, kidneys, bladder, skeleton and rest of the body. This was realized previous to image registration in a semiautomated way using the ITK-Snap toolbox. Each segmented mouse was transformed using the output transformation parameters obtained during CT image registration. The resulting segmentation was compared with the original Digimouse atlas to quantify image registration accuracy using established figures of merit such as Dice coefficient and Hausdorff distance showing fair to excellent agreement and a mean registration mismatch distance of about 6 mm. Pre-registration was applied to some PET images which were slightly misaligned with the corresponding CT images. PET images were then transformed using the same method used earlier. The results demonstrate good quantification accuracy in most regions, especially the brain. As expected, relatively large deviations were obtained for the bladder. It can be concluded that the proposed automated technique is reliable and suitable for fast quantification of preclinical PET data in large serial studies.