Blood Vessel Mapping in Retinal Images

Blood vessel detection is a critical topic in retinal image automatic processing. Blood vessel morphology is an important indicator of diabetes and hypertension. Major abnormality of blood vessel includes shape changes (diameters, smoothness, tortuosity), microaneurysm, hemorrhages, neovascular growth, loss of large vessels, etc. Most retinopathy is related to damaged vessels, and often located around vessels.

Fast and accurate mapping of the blood vessels is critical to design of automated retinopathy screening systems. We have undergone three generations of algorithm refinement, and the current version is being tested broadly in multiple clinical trials. Our algorithms have also been successfully used for a deployed laser scar detection system (slides)

Detection Outcomes (for the latest version)

The results are compared with two recently published algorithms Jiang's [6] and Staal's [7] on two widely used retinal image databases (STARE and DRIVE)

First row, the performance comparison between our algorithm and Jiang's (im0082 in STARE) [(a) our result (b) Jiang's [6] result (c) second hand-labeled ground truth].

Second row, the performance comparison between our algorithm and Staal's (00 in test images of DRIVE) [(d) our result (e) Staal's [7] result (f) first hand-labeled ground truth]

A simple review of some interesting algorithms can be found here