Detection and Classification of Diabetic Retinopathy using Histogram of Oriented Gradients and Decision Tree Classifier

Abstract

Diabetic Retinopathy is a retinal disorder commonly occurring among type-1 and type-2 diabetic patients. This paper presents a method for automated grading and classification of Diabetic Retinopathy. Depending on the pathologic abnormalities present on the retinal surface, the images are graded as Normal, Mild, Moderate and Severe. In this method, segmentation based on Morphological operations and thresholding is used to segment the regions of interest comprising microaneurysms, hemorrhages and exudates. Histogram of Oriented Gradients descriptor is generated and used as an input for Decision tree based classification. The proposed method yielded precision = 0.92, Recall = 0.93 and F-Score = 0.90. The performance evaluation measures testify the robustness of the proposed algorithm.