Scholarlink

One of the main causes of blindness is the eye disease known as Diabetic Retinopathy (DR). Diabetic Retinopathy is caused due to prolonged diabetes, so it is a chronic disease. The disease’s early detection can help to save the vision of the persons suffering with diabetes. The growth of exudates, hemorrhages, and microaneurysms shown in an eye image are clear indication of an unhealthy eye. Detection of DR can be categorized in multiple stages according to the disease progression showing features exhibited by the eye fundus images i.e. from early stage to advanced stage. In this paper, we have used two standard base deep learning models named as ResNet50V2, and VGG16 with combinations of Dense layers, Random Forest, Naïve Bayes and Logistic Regression classifiers for detection of DR in five multiple categories. In our work, we have trained and validated these models over APTOS19 dataset provided by Kaggle dataset repository. Complete tables of the comparisons of the accuracies of various combinations of the deep learning models with other classifiers are given in the result section. In our proposed architecture, the ResNet50V2 with Logistic Regression classifier has shown the 99% accuracy in training and 76% accuracy in the validation performance. It shows that, ResNet50V2 model has performed better than VGG16 model.

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

Choice of Baseline Model: Impact on Transfer Learning-Based Detection of Diabetic Retinopathy

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Subhash Chand Gupta,
  • Shripal Vijayvargiya,
  • Vandana Bhattacharjee

摘要

One of the main causes of blindness is the eye disease known as Diabetic Retinopathy (DR). Diabetic Retinopathy is caused due to prolonged diabetes, so it is a chronic disease. The disease’s early detection can help to save the vision of the persons suffering with diabetes. The growth of exudates, hemorrhages, and microaneurysms shown in an eye image are clear indication of an unhealthy eye. Detection of DR can be categorized in multiple stages according to the disease progression showing features exhibited by the eye fundus images i.e. from early stage to advanced stage. In this paper, we have used two standard base deep learning models named as ResNet50V2, and VGG16 with combinations of Dense layers, Random Forest, Naïve Bayes and Logistic Regression classifiers for detection of DR in five multiple categories. In our work, we have trained and validated these models over APTOS19 dataset provided by Kaggle dataset repository. Complete tables of the comparisons of the accuracies of various combinations of the deep learning models with other classifiers are given in the result section. In our proposed architecture, the ResNet50V2 with Logistic Regression classifier has shown the 99% accuracy in training and 76% accuracy in the validation performance. It shows that, ResNet50V2 model has performed better than VGG16 model.