AUTOMATIC DETECTION OF BACKGROUND DIABETIC RETINOPATHY DISEASE USING HYBRID MULTILEVEL THRESHOLDING AND CONVOLUTIONAL NEURAL NETWORK

Background Diabetic Retinopathy is a medical condition that causes retinal damage due to diabetes mellitus. This paper discusses a novel combination method, namely multilevel thresholding with a convolutional neural network. Multilevel thresholding resulting in disease characteristics by performing morphological vessel segmentation accurately. The segmentation process was carried out automatically by performing resizing on the grayscale image then histogram equalization. After that, the histogram equalization result was performed a Gaussian filter and Sobel edge detection to detect blood vessels. Lastly, the segmentation process was carried out using Otsu's multilevel thresholding, and then the classification phase using convolutional neural network was performed. The proposed method tested by using Stare datasets. In order to measure the performance of this technique, specifications, sensitivity, and accuracy, used. The results showed 100% specifications, 60% sensitivity and 92% accuracy using 2 level thresholding and at 3 levels of thresholding results 100% in Specifications, Sensitivity and Accuracy.