Scholarlink

Diabetic Retinopathy (DR) is one of the major causes for vision loss across the globe among diabetic people. Early detection with the help of automated analysis of retinal fundus pictures can make things easier. The systems which are being developed for the same purpose appear to be deep learning based. The paper deals with classifying the images of retina into 5 different stages of DR. This is done with the help of Convolutional Neural Networks (CNNs) and transfer learning. The data used for the starting is curated and pre-processed Kaggle data. The model is also enhanced with the help of data augmentation, use of multi-class classification and the utilization of the Grad-CAM for the purpose of the explainability. The proposed model is able to achieve an accuracy of 96.2%. Even the F1-score accuracy of 95.7%. And the AUC accuracy of 0.98 has been able to bet the all the traditional binary classifiers. The model is further enhanced to make the decision making easier with the help of the visual explanations for the same. The discussed model is one of the containers and it could be deployed in the real-time health care systems. It is definitely looking at the possibilities that the CNNs could be used to check out the conditions of each and every person depending on the fundus.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Detection of Diabetic Retinopathy Using Convolutional Neural Networks: A Deep Learning Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Karuna Gupta,
  • Shubham Kukreti,
  • Mithilesh Kumar Yadav

摘要

Diabetic Retinopathy (DR) is one of the major causes for vision loss across the globe among diabetic people. Early detection with the help of automated analysis of retinal fundus pictures can make things easier. The systems which are being developed for the same purpose appear to be deep learning based. The paper deals with classifying the images of retina into 5 different stages of DR. This is done with the help of Convolutional Neural Networks (CNNs) and transfer learning. The data used for the starting is curated and pre-processed Kaggle data. The model is also enhanced with the help of data augmentation, use of multi-class classification and the utilization of the Grad-CAM for the purpose of the explainability. The proposed model is able to achieve an accuracy of 96.2%. Even the F1-score accuracy of 95.7%. And the AUC accuracy of 0.98 has been able to bet the all the traditional binary classifiers. The model is further enhanced to make the decision making easier with the help of the visual explanations for the same. The discussed model is one of the containers and it could be deployed in the real-time health care systems. It is definitely looking at the possibilities that the CNNs could be used to check out the conditions of each and every person depending on the fundus.