Deep Learning-Based Classification of Diabetic Retinopathy: Leveraging the Power of VGG-19
Tajender Malik, Vikas Nandal, Puneet Garg
Abstract
Diabetic retinopathy (DR) is a serious eye disease that causes visual impairment and blindness if not promptly diagnosed and managed. Deep learning approaches, specifically convolutional neural networks (CNNs), have shown considerable potential for accurately classifying retinal images for DR detection. In this research paper, we propose a VGG-19-based classifier for automated DR detection and classification. By leveraging the power of the VGG-19 architecture, which is renowned for its excellent performance in image recognition tasks, our work seeks to improve the accuracy of DR diagnosis. Our dataset consists of a diverse collection of retinal fundus images obtained from patients at various stages of DR severity. We preprocessed the images to enhance their quality and extract relevant features using greyscale conversion, Gaussian filtering and circular cropping. Through transfer learning and fine-tuning, we adapt the VGG-19 model for DR classification, utilizing its learned representations to enhance our model's performance. We evaluated the classifier's performance using various parameters like, classifier accuracy, precision, recall and area under the receiver operating characteristic curve (ROC). The experimental findings show a promising accuracy of 64.5% for four-class classification and effectiveness of our VGG-19-based classifier in identifying four stages of DR. Our proposed classifier holds significant potential for assisting healthcare professionals with early diagnosis and intervention. In conclusion, this research paper presents an innovative VGG-19-based classifier for automated DR detection, offering a valuable tool for efficient and accurate DR screening and diagnosis.