Scapula Statistical Shape Model construction based on watershed segmentation and elastic registration

Automated bone segmentation is one of the most challenging problems in medical imaging. The increasingly demanded MR imaging suffers from low contrast and signal-to-noise ratio when it comes to bones. To increase the segmentation robustness, a prior model of the structure could guide the segmentation when explicit information is missing or weakly presented. Statistical Shape Models (SSMs) are efficient examples for such application where a set of dense correspondences between the training samples is to be established. The complexity of the anatomy of the scapula's bone is a real challenge at this level. We present an automated SSM construction approach with an adapted initialization to address the correspondences problem. Our approach is atlas-based where landmarks are matched on each sample using rigid and elastic registration. Our innovation stems from the derivation of a robust SSM based on Watershed segmentation which steers the elastic registration at some critical zones.