Diabetic retinopathy detection by optimized deep learning model

In the medical image analysis, the diagnosis of diabetic retinopathy (DR) from fundus images are identified as an open challenge and requires possible solutions. The major stages of the proposed DR are Pre-processing, Segmentation, Feature Extraction, and Classification. In Pre-processing, the retinal fundus images are RGB images, among them the G-channel is selected. Following that, histogram equalization and contrast limited adaptive histogram equalization (HE and CLAHE) are applied. Then the next stage is removing the optic disc (OD) and it is done by Circle Hough Transform (CHT). Then, the Gray Level thresholding is used for removing the blood vessels. Then the Exudates are segmented by the Modified Expectation Maximization (MEM) algorithm. Then Gray Level Co-occurrence Matrix (GLCM) is used for feature extraction. At last, features are classified by the Deep Neural Network with a Butterfly Optimization Algorithm (DNN-BOA) classifier which is used for classifying the several stages of DR. The proposed scheme is implemented on MATLAB 2021a. The performance of the implemented of the proposed scheme is compared with the other approaches with some measures like precision, accuracy, sensitivity, F-score and specificity on the DIARETDB1 and MESSIDOR datasets. The accuracy of the proposed scheme is 0.983 and 0.989 on the two datasets respectively. The accuracy of the proposed scheme is 25.9%, 23.29%, 14.5% and 16.6% better than the approaches like KNN, SVM, DNN and DBN on the MESSIDOR dataset.