Automatic Aortic Root Segmentation with Shape Constraints and Mesh Regularisation

A non-invasive procedure called Transcatheter Aortic Valve Implantation (TAVI) has emerged as an alternative procedure for patients suffering with aortic stenosis, but cannot undergo standard open-heart surgery. A full segmentation of the aortic root is important to the success of the procedure, and is essential for patient selection, procedural planning, and post-evaluation [1]. We propose a fully-automatic, deformable model-based method to segment the aortic root in 3D cardiac CT images. This consists of aligning an initial mesh with an initial aortic root pose estimation, before deforming the mesh towards the object boundary in the image. The estimation of the aortic root pose can be considered as an object detection problem, and a marginal space learning (MSL) method is adopted from [3] for this purpose. Once the initial mesh is aligned with the estimated pose, we implement a two-stage mesh deformation method: non-iterative boundary segmentation followed by iterative boundary refinement with mesh smoothing. Figure 1 outlines the steps taken at the testing stage for our automatic segmentation method.