Revolutionizing Diabetic Retinopathy Diagnosis: Harnessing Automated Deep Learning Classifiers for Early Detection

Abstract

Diabetic Retinopathy (DR) is a prevalent microvascular complication of diabetes, often leading to vision impairment if not diagnosed and managed promptly. This paper presents an innovative approach for the early detection of DR utilizing automated deep learning classifiers. The proposed methodology integrates advanced image processing techniques with state-of-the-art deep convolutional neural networks (CNNs), aiming to achieve accurate and efficient DR detection. The initial stage of image enhancement involves the application of a Median filter to mitigate noise and enhance the quality of retinal fundus images. Subsequently, the segmentation process utilizes Fuzzy C Means (FCM) to partition the images into meaningful regions, aiding in the isolation of potential pathological lesions associated with DR. Texture features are then extracted using the Gray-Level Co-occurrence Matrix (GLCM) technique, capturing crucial textural information from the segmented regions. The extracted features undergo further processing through a deep CNN, which serves as a powerful classifier. The deep CNN is meticulously designed and trained to learn intricate patterns and representations in the feature space. By leveraging a substantial amount of labeled data, the classifier demonstrates a high degree of sensitivity and specificity in identifying early signs of DR. The amalgamation of preprocessing, segmentation, feature extraction, and deep CNN-based classification culminates in an efficient and reliable tool for early DR detection and achieves accuracy of 98.67%. © 2023 IEEE.