Automated Detection of Diabetic Retinopathy Images using Pre-trained Convolutional Neural Network

Diabetic retinopathy is one of the most common retinal diseases in diabetic patients, resulting in avoidable blindness. Therefore, early detection and grading of retinal images are crucial to reduce the risk of vision loss. Also, the early detection of DR in an automated manner is critical for effective care. The objective of this study is to classify and evaluate the retinal fundus images into two levels. In the current scenario, compared to feature-based image classification techniques, convolutional neural networks have a higher image classification efficiency. This work presents a comparative evaluation of four pre-trained models for binary classification of fundus images. The evaluation output indicates that the Xception model outperforms all the three models with an accuracy of 99.3 %. The efficacy of the pre-trained model is evaluated by comparing the sensitivity, specificity, precision, F-score, and quality index of each model.