Fine tuned vgg19 convolutional neural network architecture for diabetic retinopathy diagnosis

Abstract

Diabetic Retinopathy (DR) is an eye-related complication experienced by individuals with longstanding diabetes. Usually diagnosed by the healthcare professional by retinal fundus examination during Medical check-ups or mass screenings. Early detection of diabetic retinopathy will avoid vision loss and other issues. The objective of this work is to diagnose the Diabetic Retinopathy from retinal fundus images using deep learning (DL) techniques for better detection accuracy.The proposed fine-tunedVGG19 CNN architecture is performing well with the Kaggle data set and effectively dealing with this multi-class classification problem. The proposed model uses pre-trained weights from image net data which lessens the training time and improves the performance in detecting DR from retinal fundus images in terms of sensitivity, specificity, and accuracy. Deep transfer learning with fine-tuning method implementation was carried out to get the highest test accuracy of 73.60%.