Automated Image Analysis Of Fundus Photographs For Mass Health Screenings

As the use of non-mydriatic wideangle fundus cameras has become widespread, a large number of color fundus photographs are now routinely taken during group health checks. Since many tests have to be performed at one time on a large number of people during group physical examinations, a utomation or simplification of these tests is important. Bearing this in mind, we recently experimented with an automated image analysis of color fundus p hotographs (taken during group health checks) with a low-cost s ystem using a personal computer. The following results were obtained: (1) By moving regression planes, corresponding to minute segments of the image, vascular edges could be detected without any limitation as to the direction of approach. (2) Data transformation into Y.1.G signals with the NTSC (National Television System C ommittee) method allowed for e ffective e xtraction of vascular patterns without being affected by color changes. (3) Chain c oding of the extracted vascular patterns r esulted in reduction of the data to about 2% of the original