Computer Aided Diagnosis of Diabetic Retinopathy Using Multi-Resolution Analysis and Feature Ranking Frame Work

Diabetes is a rapidly increasing worldwide problem characterized by impaired metabolism of glucose causing long-term dysfunction and failure of various organs. Diabetic Retinopathy (DR) is one of the most common complications of diabetes and is also one of the major causes of blindness and visual impairment. Retinal fundus photography is widely used in the diagnosis of various eye diseases including DR and automatic retinal image analysis is emerging as an important screening tool for early disease detection. In this work, we have extracted features from Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT) coefficients. The features were ranked using Entropy, Chernoff Bound and Bhattacharya Distance (CBBD), minimum Redundancy Maximum Relevance (mRMR)-Mutual Information Difference (MID) and mRMR-Mutual Information Quotient (MIQ) methods in order to select best ranking scheme and optimum features for discriminating between normal and DR retinal images using supervised classifiers. Then the ranked features were fed to AdaBoost for automated diagnosis. Our results show that, the proposed system was able to yield an average accuracy of 94.17%, sensitivity of 92.81%, and specificity of 96.27% using twelve features selected using mRMR-MID feature ranking method and AdaBoost-Least-squares (LS) learning method. The proposed system can potentially be used as part of a community-based automatic mass DR screening program.