In the realm of medical image analysis, particularly aorta segmentation, non-rigid registration stands out as a crucial preprocessing step. The application of tools such as ITK (Insight Segmentation and Registration Toolkit) is frequently suggested to achieve this alignment. While lacking prior experience with ITK, its potential for integration within platforms like Slicer3D is acknowledged, alongside the observation that despite predominantly C++-based examples, Python accessibility via pip-install within Slicer facilitates its use. Resources detailing the utilization of ITK filters in Python via SimpleITK are also noted as valuable starting points.
Driven by curiosity and leveraging freely available time, the creation of a Statistical Shape Model (SSM) for vertebrae from the VerSe dataset is being explored. This initiative aims to investigate the efficacy of SSMs in achieving robust segmentation outcomes. Collaboration and result sharing are encouraged, with the recognition that a bounding box detection algorithm would significantly benefit segmentation processes, especially when combined with active shape model methodologies.
An innovative concept under consideration involves leveraging SSMs for automatic landmarking. The hypothesis posits that by marking landmarks on the mean model and models incorporating eigenvectors within an SSM—whether for vertebrae or other anatomical structures—landmark positions can be derived for any registration performed. This approach holds promise for automating the landmarking process.
Collaboration is actively sought to test this hypothesis. Should an SSM be developed prior to others, offers of assistance are extended to validate this landmarking concept.
Mauro
