Automated Glaucoma Detection using Machine Learning Approaches

Abstract

One among the most common reasons for people surviving with eye disorders or visual impairment around the globe is glaucoma. It is one of the prime roots that is leading to irrevocable blindness in our world today. Manual examination of illness by medical practitioners today do not guarantee precision. Therefore early detection of such flaw in eye is a rising need today and this can be aided by automation of glaucoma detection. Input images used in this proposed work are extracted from DRIVE database that comprises of both images belonging to glaucoma group (abnormal) and normal images. These images Pre-processed by first converting them to RGB (to know which channel among R,G,B can give the highest contrast). Binarization of an image is performed to obtain a BW (black and white) image followed by dilation which is a  morphological operation that uses a structuring element to look into the shape   and expanding the input image by filling holes and making it more visible. Optic disc (starting of optic nerve) and optic cup (depression in nerve centre) which are terms of interest for glaucoma detection are made clearer. CCL is applied to find ROI which is the optic cup part of the disc here. For feature extraction, cup to disc and rim to disc ratios are extracted. A classifier is built using SVM that gave an overall accuracy of 88% for GLCM and 97% for CDR..