Enhancing Diabetic Retinopathy Image Classification using CNN, Resnet, and Googlenet Models with Z-Score Normalization and GLCM Feature Extraction

If diabetic retinopathy is not identified and treated right away, it is a dangerous side effect of diabetes that will cause blindness. In this work, we investigate novel deep learning (DL) techniques for the automated categorization of DR images. We examine the efficiency of three powerful convolutional neural network (CNN) frameworks: the basic CNN, ResNet, and GoogLeNet. We employ preprocessing techniques as Z-score normalization for pixel brightness leveling and Gray Level Co-occurrence Matrix feature extraction for gathering in-depth texture information in order to increase the models' accuracy. Our tests show that CNN outperforms alternative models in DR image categorization, particularly when enhanced with ResNet and GoogLeNet. According to experimental investigations, the recommended work produced outcomes with a 94% accuracy rate, a 0.92 sensitivity rate, and a 0.90 specificity rate. The tool used for execution Jupyter Notebook and language used is python.