Multi-Class Classification for Prediction of Retinal Diseases (Retinopathy and Occlusion) from Fundus Images

Abstract

Medical Imaging and Mining provide very useful information for the medical practitioners. In the field of ophthalmology, fundus images capture the retina of the eye. A lot of diseases like different types of retinopathy, types of occlusion, Choroidal Neo-vascularisation, Glaucoma, Macular Degeneration etc., can be diagnosed from the analysis of the fundus retinal images. Retinopathy and Occlusion are the sight threatening diseases that demand early detection. In this paper, prediction of Retinopathy, Occlusion, other disease affected and normal cases is attempted through extraction of overall image features. The automatic detection is done through color channel extraction, contrast enhancement, overall image feature (Statistical, GLCM, Histogram based features) extraction and classification. STARE, a publicly available repository of retinal fundus images is used for training the system. It is observed that Random Tree Classifier yielded the best performance achieving an accuracy of 96.15% in detecting presence of disease, 76.92% in detecting Retinopathy, affected by other disease or normal and Random Committee classifier yielded the best performance yielding an accuracy of 84.62% in detecting Occlusion, affected by other disease or normal and 64.12% in detecting Retinopathy, Occlusion, affected by other disease or normal.