Retinal image registration using topological vascular tree segmentation and bifurcation structures

Abstract

This paper presents a new retinal image segmentation and registration approaches. The contribution of this paper is two-fold. First, the conventional vessel-tracking methods use local sequential searching, which can be easily trapped by local intensity discontinuity or vessel rupture. The proposed method uses global graph-based decision that can segment the topological vascular tree with 1-pixel width and fully connection from retinal images. Staring from initial multi-scale ridge segmentation, the disconnected vessels are retrospectively connected and then spurious ridges are removed using a shortest path algorithm on a specially defined graph. The hypothesis testing is defined in terms of probability of pixel belong to foreground and background, which enables that the false detections could be removed. Second, the conventional point-matching methods largely depend on the branching angles of single bifurcation point. The feature correspondence across two images may not be unique due to the similar angle values. In view of this, structure-matching registration is favored. The bifurcation structure is composed of a master bifurcation point and its three connected neighboring pixels or vessel segments. The characteristic vector of each bifurcation structure consists of the normalized branching angle and length, which is fairly robust to be against translation, rotation, scaling, and even modest distortion. The experimental results are presented to demonstrate the superior performance of the proposed approach.