Diabetic Retinopathy Detection Using Soft Computing Techniques

Diabetes, referred to as diabetes mellitus, describes a group of metabolic diseases in which the person has high blood glucose. Possible complications that can be caused by badly controlled diabetes: Eye complications, Foot complication, Skin complications, Heart problems, Hypertension, etc. Diabetic retinopathy (DR), a common complication of diabetes, affects the blood vessels in the retina. It is due to the retina not receiving enough oxygen. In this work, publically available MESSIDOR database with 1200 fundus images are considered. Due to the textural changes in the retinal fundus images texture analysis methods like statistical moments and GLCM are used and extracted 40 features. The unbalanced dataset decreases the accuracy so oversampling is done before classification. Classifiers such as Support Vector Machine (SVM), Random Forests, Gradient boost, AdaBoost, Gaussian Naive Bayes used to detect the DR. Each algorithm has its own advantages and varying accuracy, SVM compared to others have good accuracy of 88.71%, Random forests technique accuracy of 83.34%, Gradient boost algorithm accuracy of 83.34%, while AdaBoost and Gaussian Naive Bayes methods reached accuracy of 54.3% and 37.09% respectively.