Automatic Optic Disc segmentation using maximum intensity variation

Abstract

Optic disc (OD) segmentation is an important step in automating eye screening for pathological conditions. In this paper, we propose an intensity-based approach to detect the OD boundary, given OD center. OD center is utilized to crop the sub-image that encloses the OD, within which candidate contour points are obtained. Points of maximum intensity variation, both horizontally and vertically, are chosen as candidate contour points. Iterative curve fitting is carried out, incorporating smoothness constraints. The area within the contour is checked for values of mean intensity, variance and compactness. The algorithm is applied on 152 images taken from two public datasets, DIARETDB1 and MESSIDOR. The validation criteria used are common area overlapping between automated segmentation and true OD region (score), sensitivity and Mean Distance to Closest Point (MDCP). The algorithm renders, on an average, a score value of 90%, sensitivity of 93% and MDCP of 8.3 pixels.