Feature Extraction and Analysis for Diabetic Retinopathy Classification using Pre-trained Deep Networks

In the quest to improve the accuracy of diagnosing diabetic retinopathy, we explore a novel approach—leveraging pre-trained deep networks for feature extraction. This method involves harnessing the knowledge acquired by neural networks through prior training on diverse datasets. By utilizing these pre-trained networks, we aim to extract intricate features from retinal images, crucial for discerning the stages of diabetic retinopathy. This approach not only capitalizes on the depth of information learned by the networks but also enhances the efficiency of feature extraction. Our study delves into the application of this technique, evaluating its performance in diabetic retinopathy classification. Through sophisticated technology and medical insight, we seek to unveil a more nuanced and accurate means of identifying and classifying diabetic retinopathy, paving the way for improved patient care and timely interventions. The MobileNet V2 has obtained the highest classification accuracy of 92.42% over the other considered pre-trained models through the significant feature extraction.