Computer Aided Hierarchal Lesion Classification for Diabetic Retinopathy Abnormalities

Abstract

Vision loss from Diabetic Retinopathy (DR) abnormalities can be prevented by employing timely treatment and continuous monitoring of disease progress. Early diagnosis can effectively expedite the success rate of disease curability. Automated computer aided diagnostic systems can aid the ophthalmologists and prevent their tedious and time consuming efforts using manual lesion detection approaches. Computer Aided Hierarchal Lesion (CAHL) classification approach is proposed in this paper utilizing optimal classifiers with optimal feature set for early and efficient DR diagnosis. Exhaustive statistical investigation of extracted shape and intensity features resulted in prominent features which were used for abnormality classification employing SVM, kNN and NN classifiers. The proposed CAHL approach achieved best classification performance for NN classifier in terms of four statistical indices: accuracy, sensitivity, specificity, positive prediction value of 100% for both normal and abnormal stage classification as well as DR abnormality classification. A trade-off between run-time and high cost of manual computation is maintained using NN classifier based mechanism for DR classification. The proposed method outperforms the state of the art techniques when compared to the recently published methods for DR screening. Critical DR problems like neovascularisation and blood vessel bleeding will be addressed in the future part of the research.