Detección automática del disco óptico en retinografías digitales patológicas.

Abstract

In the current work, machine vision techniques for optic disc (OD) detection in medical images with fundus pathology are studied. This study includes the analysis of images from four public image sets used in other scientific investigations. Sharpness, contrast, contrast limited adaptive histogram equalization (CLAHE) corrections are implemented. Finally, the Entropy metric is run to determine according to its result, the location of the optical disk in a candidate block. Two methods of OD localization are proposed that will allow to locate the optical disc quantitatively in the two cases. The first method will locate the OD in two stages by dividing the image into 9 blocks and sliding windows; the block with the highest entropy will be the one where the optical disk is located. The second method will locate the OD directly through five sliding windows; the candidate block will be the one with the highest entropy and that is where the OD will be located. The tests allow to verify that the Entropy metric provides results that allow to locate, in a quantitative way, the optical disk of most of the images; the best results for the first method were 96.08% in Drishti-GS Test and for the second method were 91.84% in DrishtiGS Training. The Entropy metric is applied to each region of the image. The block size has been previously determined, and it can be experimentally concluded that the best results are obtained with nine blocks of equal size along the analyzed image.