Fundus Image Screening for Diabetic Retinopathy

Abstract

The proposed system aims in diagnosing Diabetic Retinopathy (DR), a snag in patients with Diabetes for prolonged periods. MinIMaS algorithm is primarily used for background Extraction and Gaussian Mixture Model (GMM) is used for grouping of lesions. The amount of false positives are minimized by the selection of feature vectors used for classification. The proposed system achieves sensitivity of 89% for grouping of bright lesion and sensitivity of 82% for classifying bright lesion which when compared to existing system (KNN classifier) gives a higher sensitivity value. Similarly MinIMaS algorithm provides 91% accurate results when compared to the existing Haar based wavelet transform (70%) and Highest Average Variation (65%). System aids in detecting severity of Diabetic Retinopthy with improved accuracy for timely treatment of patients.