Segmentation of vessels using weighted local variances and an active contour model

Vessel segmentation plays an important role in diagnosing patients with vascu-lar problems. Vascular segmentation refers to the distinguishing of vasculature from background regions in images. Performing segmentation of vasculature with blurry and low contrast boundaries in noisy images is a challenging problem. The main reason is that low contrast and noise reduces recognizibility of vasculature. It is observed that traditional operations, like low pass filtering, is not suitable in this situation. This is because such operations annhilate narrow objects such as vascular structures. This thesis presents a novel approach on segmenting blood vessels using weighted local variances and an active contour model. In this work, object boundary orientation is estimated locally based on the calculation of weighted local vari-ance. Such estimation is less sensitive to noise compared with other common approaches. Edge clearness is measured by the relation of weighted local vari-ances obtained along different orientations. The formulation of the proposed method is independent of the edge intensity contrast and capable of locating weak boundaries. Integrating the orientation and clearness of edges, an active contour model is employed to align contours that match the contour tangent direction and edge orientation. Contours of the active contour model is based on a level set frame-work, which can handle the merging or splitting of contours naturally. With such ability, segmenting vessels using the proposed method can either be fully automatic, or involve manual interactions in order to permit users to specify the target vessels. The proposed method is validated by three synthetic images and nine clin-ical cases. It is experimentally shown that our method is suitable for dealing with noisy images which consist of structures having blurry and low contrast boundaries, such as blood vessels