Deep Feature Extraction and Classification of Diabetic Retinopathy Using AlexNet, InceptionV3, and VGG16 CNN Architectures

Diabetic retinopathy (DR) is a significant cause of vision impairment and blindness among diabetic patients, characterized by progressive retinal damage. Early and accurate detection is crucial for effective management and treatment. This research explores advanced deep learning techniques to enhance DR detection and classification by leveraging Convolutional Neural Networks (CNNs). We propose a novel methodology incorporating deep feature extraction and classification using three CNN architectures: AlexNet, InceptionV3, and VGG16. Our approach involves extracting deep features from retinal images to capture intricate patterns associated with various DR stages, followed by classification to differentiate between healthy and various stages of DR. The dataset used include publicly available Fundus Image Registration Dataset (FIRE) for comprehensive evaluation. Detailed preprocessing steps ensured data quality and relevance, while feature extraction techniques harnessed the strengths of the selected CNN architectures. The performance of the proposed models was evaluated based on accuracy, sensitivity, precision, and F1-score. Our results demonstrate that AlexNet achieves the highest accuracy at 95.37%, outperforming InceptionV3 and VGG16. This study underscores the effectiveness of CNN-based approaches in DR detection and highlights the potential for further improvements in early diagnosis and treatment strategies.