Segmentation of Abdominal Aortic Aneurysms with a Non-Parametric Appearance Model

Abstract

This paper presents a new method to segment abdominal aortic aneurysms from CT angiography scans. The outer contour of lumen and thrombus are delineated with independent 3D deformable models. First the lumen is segmented based on two user indicated positions, and then the resulting surface is used to initialize the automated thrombus segmentation method. For the lumen, the image-derived deformation term is based on a simple grey level appearance model, while, for the thrombus, appearance is modelled with a non-parametric pattern classification technique (k-nearest neighbours). The intensity profile along the surface normal is used as classification feature. Manual segmentations are used for training the classifier: samples are collected inside, outside and at the given boundary. During deformation, the method determines the most likely class corresponding to the intensity profile at each vertex. A vertex is pushed outwards when the class is inside; inwards when the class is outside; and no deformation occurs when the class is boundary. Results of a preliminary evaluation study on 9 scans show the method’s behaviour with respect to the number of neighbours used for classification and to the distance for collecting inside and outside samples.