Retinal network characterization through fundus image processing: Significant point identification on vessel centerline

This paper describes a new approach for significant point identification on vessel centerline. Significant points such as bifurcations and crossovers are able to define and characterize the retinal vascular network. In particular, hit-or-miss transformation is used to detect terminal, bifurcation and simple crossing points but a post-processing stage is needed to identify complex intersections. This stage focuses on the idea that the intersection of two vessels creates a sort of close loop formed by the vessels and this effect can be used to differentiate a bifurcation from a crossover. Experimental results show quantitative improvements by increasing the number of true positives and reducing the false positives and negatives in the significant point detection when the proposed method is compared with another state-of-the-art work. A sensitivity equal to 1 and a predictive positive value of 0.908 was achieved in the analyzed cases. Hit-or-miss transformation must be applied on a binary skeleton image. Therefore, a method to extract the vessel skeleton in a direct way is also proposed. Although the identification of the significant points of the retinal tree can be useful by itself for multiple applications such as biometrics and image registration, this paper presents an algorithm that makes use of the significant points to measure the bifurcation angles of the retinal network which can be related to cardiovascular risk determination.