Improving the Accuracy of Diabetic Retinopathy Severity Classification with Transfer Learning
Diabetic Retinopathy (DR) is a major cause of blindness in Diabetic patients, and its early detection benefits diagnosis and subsequent treatment methods. In this work, a convolutional neural network uses the VGG-16 model as a pre-trained neural network for fine-tuning, and, thereby classifying the severity of DR. The model also uses efficient deep learning techniques including data augmentation, batch normalization, dropout layers and learn-rate scheduling on high resolution images to achieve higher levels of accuracy. An average class accuracy (ACA) of 74%, sensitivity of 80% at a specificity of 65% and area under the curve (AUC) of 0.80 have been achieved, which are higher than previously reported results obtained using other pre-trained networks or models.
Paper
Full text
Improving the Accuracy of Diabetic Retinopathy Severity Classification with Transfer Learning
Semantic Scholar · Medicine · 2020
Abstract
Diabetic Retinopathy (DR) is a major cause of blindness in Diabetic patients, and its early detection benefits diagnosis and subsequent treatment methods. In this work, a convolutional neural network uses the VGG-16 model as a pre-trained neural network for fine-tuning, and, thereby classifying the severity of DR. The model also uses efficient deep learning techniques including data augmentation, batch normalization, dropout layers and learn-rate scheduling on high resolution images to achieve higher levels of accuracy. An average class accuracy (ACA) of 74%, sensitivity of 80% at a specificity of 65% and area under the curve (AUC) of 0.80 have been achieved, which are higher than previously reported results obtained using other pre-trained networks or models.