Proposal of integrated analysis system for retinal fundus images and prototyping of an interactive classification tool for retinal patterns

Abstract

We proposed a three-dimensional quantitative analysis tool for posterior shapes of eyeballs deformed by such as high myopia diseases, using 3D-MRI. In order to analyze temporal progressions of this disease, we have also proposed developing an integrated vessel analysis system for retinal fundus images. In this system, an extraction tool of retinal vessels, a quantitative analysis tool of retinal vessels and a registration tool of retinal images have been implemented. Using these tools we could register given retinal image with past images shot about 20–30 years ago, and make comparative analysis between time-series retinal images. However, precision of vessel extraction was not enough for making the registration process automatic, and it was difficult to analyze temporal deformation of individual artery or vein with extracted vessel patterns. In this paper, we propose an interactive classification tool for retinal patterns, which is composed of an automatic extraction tool of retinal vessels and optic discs, an automatic separation tool of arteries and veins, and an editing tool of segmented images. Using this newly proposed tool, we can obtain a precisely segmented retinal image, which is divided to seven areas as artery, vein, artery-vein crossover, optic disc, cup, PPA and other parts. Although this tool includes manual operations, we are considering decreasing load of operations by implementing a machine-learning function to the editing tool.