Transfer Learning Based Approach for Diabetic Retinopathy Classification using Fundus Images

Diabetic retinopathy (DR) is a major microvascular complication of diabetes. Around 95 million individuals worldwide suffer from DR. Regular testing of fundus images and early identification of initial diabetic retinopathy symptoms, namely microaneurysms and hemorrhages, are essential to decrease vision impairment possibilities. This research work is focused on the detection and classification of fundus images of diabetic retinopathy. In this research work, we have proposed a deep learning-based method to classify diabetic retinopathy fundus images into positive (diabetic) class and negative (normal) class. The convolutional neural network is recently most popular in the computer vision for pattern recognition and classification. In this work we have used pre-trained ResNet50 for the fundus image classification. ResNet50 has amazing power to extract robust and discriminating features from the images for diagnosis. The evaluate the performances of the proposed approach we use publically available Messidor dataset. The proposed approach achieves accuracy of 91.78 % and sensitivity of 94.68 %.