A Retinal Vessel Tracking Algorithm Based On Bayesian Method

Retinal vascular network is the only human blood microcirculation that can be observed by noninvasive ways. Its characteristics are essential for clinical diagnosis of many diseases, such as arteriosclerosis and diabetes. Automatic segmentation of retinal vessels in fundus images is one of the fundamental techniques in computer aided diagnosis systems. It is also of great importance in mass screening and diagnosis of relevant diseases automatically.Retinal vessel segmentation has been investigated by many researchers. Although the existing methods can be categorized into many groups, they all depend on similar characteristics of retinal vessels. Vessel tracking approach can provide more information of each vessel than pixel-based method. A retinal vessel boundary tracking method based on Bayesian theory and multi-scale line detection is proposed in this paper. The proposed method can segment retinal vessel network automatically. Meanwhile, it can provide information about vessel structure and the change of vessel width.To make the proposed algorithm fully automatic, initial points should be determined. Retinal vessels have an important characteristic that all of them originate from a structure called optic disk. So the first step is to locate the optic disk and specify some initial conditions so that manual intervention is not necessary. In our algorithm, vessel structures are classified into three categories: normal vessel, branching and crossing. According to the fact that the direction of a vessel will not change too much in a small distance, multi-scale line templates are applied on each candidate point, and the responses are transformed into a probability form. Meanwhile, our algorithm uses Gaussian model to match the intensity distribution of vessels along the cross section. The main contribution of the proposed method is that it takes the characteristics of vessel point in both dimensions into consideration, which extends the 1-D vessel model to 2-D vessel model.The REVIEW database is used to validate our method and comparison study is performed with several classic and state-of–the-art methods. The result shows that our algorithm can provide the information about vessel width successfully in most cases. Moreover, it is robust and has good performance in overcoming central light reflection effect. In some image database, our method has the best performance.