ULTRASOUND INTERACTIVE SEGMENTATION WITH TENSOR-GRAPH METHODS

We address the problem of segmenting aortic aneurysms in ultrasound images. As solution we propose a novel framework based on graph-based interactive segmentation methods, such as graph-cuts and random walks. Our main contribution is extending these approaches to handle structure tensor ultrasound images. Our hypothesis is that the structure tensor is better suited to represent the contextual information in ultrasound images than the pure b-mode intensity values. We demonstrate that this extension significantly improves the performance of both methods in clinical data.