Automatic optic disc segmentation using a 3D Morphologic Fuzzy Histon Roughness Index

Abstract

Fundus images have become increasingly important due to the growing number of diabetics and their risk of developing diabetic retinopathy. Automated diagnosis of retinal images requires the separation and analysis of retinal elements. Among these elements, the optic nerve head is the most significant since it is used as a reference for detecting other anatomical structures such as the fovea and blood vessels. For this reason, optic disc segmentation is essential. This paper presents a method for locating and segmenting the optic disc using the circular Hough transform and the roughness index, respectively. We used a location algorithm based on the multiscale energy filter for fast location and the medial Hough transform as the decision criterion. Segmentation was performed with Atanassov’s Intuitionistic Fuzzy Set representation and an algorithm using the 3D roughness index and mathematical morphology to smoothly segmenting. This approach to automatic segmentation was tested on the MESSIDOR database, achieving 99.59% accuracy and 91.56% sensitivity. Other types of roughness indices were also evaluated; however, color correlation within the color space of the object itself produced better results for the proposed algorithm. We hope these findings contribute to the development of reliable automatic diagnostic tools for retinal images.