A retinal vessel tracking method based on Bayesian theory

Abstract

A vessel tracking approach using maximum a posterior probability is investigated in this paper. The optic disk is detected automatically using PCA method. The Gaussian filter and intensity-gradient co-occurrence matrix are employed to segment retinal vessel. The starting points of vessels are detected around the optic disk based on the segmentation results. For each vessel, vessel tracking is performed using Bayesian theory. A semi-ellipse is defined as a searching region according to the current vessel's width, travel direction, and curvature. Candidates of next vessel edge points are selected on the semiellipse. Three vessel structures are considered: normal vessel, vessel branching, and vessel crossing. At each step, the probabilities of all combination of candidate points are calculated and vessel structure and corresponding vessel edge points are determined via Bayesian theory with the maximum a posterior. In our approach, the starting points of vessel tracking can be detected automatically. The setting of probability calculation is revised to strengthen the local linearity of retinal vessel. Our experimental results show that our proposed method can achieve satisfactory tracking results.