A variational Shearlet-based model for aortic stent detection

In medical applications, stent segmentation in the abdominal aorta has to be carried out in challenging conditions, since one has to deal with noise, low contrast, objects having similar appearances and missing or blurred edges. Variational segmentation methods eases this task by carrying prior information on the target region or on the regularity of its boundaries. In this paper, we propose a new approach based on the global minimization of the Active Contour model using the L¹-norm of the Shearlet Transform instead of Total Variation (TV -norm). One of the distinctive features of such a regularization is that it allows the detection of anisotropic structures in images like stents boundaries. The sparsity imposed by the minimization provides piecewise smooth solutions with C²-singularities. We also use the shearlet coefficients to construct an edge function for more faithful contour detection. Performances of our algorithm are evaluated on a stent segmentation from post-operative CT data. Results show that the proposed method drastically improves the detection of the stent placement compared to the TV based approach.