Classification of Diabetic Retinopathy and Normal Retinal Images using CNN and SVM

Diabetic retinopathy is a disease caused by chronic diabetes and can cause blindness. Therefore early detection of diabetic retinopathy is essential to prevent the increased severity. An automated system can help detect diabetic retinopathy quickly for determining the follow-up treatment to avoid further damage to the retina. This study proposes a deep learning method for extracting features and classification using a support vector machine. We use the high-level features of the last fully connected layer based on transfer learning from Convolutional Neural Network (CNN) as the input features for classification using the support vector machine (SVM). This method reduces the computation time required by the classification process using CNN with fine-tuning. The proposed method is tested using 77 and 70 retinal images from Messidor database of base 12 and base 13 respectively. From the results of the experiments, the highest accuracy values are 95.83% and 95.24% for base 12 and base 13 respectively.