Efficient Algorithm for Optic Disc Segmentation in Retinal Images.

Abstract

Segmentation of the optic disc from retinal fundus images is a crucial step within various algorithms aimed at detecting eye pathologies like glaucoma and diabetic retinopathy. Accurate segmentation of the optic disc is vital for the automatic detection of such pathologies. In this study, we present a simple yet highly effective approach for segmenting optic discs from retinal fundus images. Our objective is to provide an uncomplicated yet robust segmentation method that can be executed swiftly. We propose an algorithm that leverages fundamental image processing principles, specifically thresholding in conjunction with morphological operations, to accomplish optic disc segmentation. Despite its reliance on basic image processing techniques, our algorithm yields exceptional effectiveness while demanding minimal computational time. The evaluation of our approach utilizes the well-known publicly available Drishti-GS dataset. The evaluation of our segmentation algorithm produced compelling results, achieving a Sørensen-Dice coefficient of 93.95%, a Jaccard index of 88.9%, precision of 99.67%, recall of 99.89%, and a Matthews Correlation Coefficient of 94%. Additionally, the accuracy of the localized centers within the final segmentation masks was measured using the Euclidean distance, resulting in an accuracy of 99.67%. Notably, our algorithm was executed on the complete Drishti-GS dataset, encompassing 101 images. This study introduces a straightforward yet highly effective approach for optic disc segmentation in retinal fundus images. The algorithm’s reliance on basic image processing operations does not compromise its performance, as evidenced by the exceptional evaluation metrics achieved. These outcomes underscore the significance of our proposed method in facilitating the automatic detection of various eye pathologies. The efficiency of our approach, along with its promising results, suggests its potential application in broader medical image analysis contexts.