Detection And Classification Of Diabetic Retinopathy Using Adaptive Boosting And Artificial Neural Network

The disorders identified with retina of the eye such as: Diabetic Retinopathy (DR), age-related Macular Degeneration (AMD), Glaucoma, etc. can cause visual impairments. The retinal fundus pictures of the patients can be acquired with a computerized fundus camera. It can then be used as machine learning application against the manual technique for the detection and prevention of DR. In this research, the retinal fundus image obtained from Himalaya Eye Hospital Pokhara, MESSIDOR database and EyePACS are used for the detection of DR along with its severity label. In the preprocessing phase, the images are first converted to grayscale images. Histogram equalization technique is performed to adjust image intensity in order to enhance contrast. Median filtering, a non-linear digital filtering technique, is implemented to remove noise present in the image. It too helps to improve the result for further processing. Blood vessels are detected and optic discs are identified and removed. Exudates are segmented for further processing. Finally, selected features are extracted for classification. The classification by Adaptive Boosting (AdaBoost) classifier using selected features achieved precision: 0.642, sensitivity: 0.870, accuracy: 0.620 and F1-score: 0.739. While the Artificial Neural Network achieved an accuracy of 98.43% for the detection of Diabetic retinopathy and 84.21% for the severity classification as stage 0: No-DR, 1: Mild-DR, 2: Moderate-DR, 3: Severe-DR and 4: Proferative-DR.