Scholarlink

This study explores the ResNet-50 model for classifying fundus camera images into severity stages of diabetic retinopathy (DR). This study resolves the issue of limited access to medical specialists in rural regions by improving the speed and accuracy of DR detection. DR was classified into five severity levels, including No DR, Mild, Moderate, Severe, and Proliferative DR, using a dataset containing 1928 test images and 3363 training images. Image augmentation techniques were applied to improve generalization and model performance. ResNet-50 outperformed other architectures, including DenseNet, VGG16, AlexNet, InceptionV3, and classical CNN, with a specificity of 94.8%; because of its efficient utilization of residual connections to overcome the gradient-vanishing venture, ResNet-50 permitted deep learning and up-to-the-minute feature extraction. The study successfully identified the DR data without compromising computational efficiency and showed decent reproducibility for DR identification. These results suggest that ResNet-50-based models are a potential approach for automated DR severity classification, which could improve early diagnosis and management of the disease in resource-poor settings.

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 ResNet-50 Deep Learning Model

  • 章节

  • 发表日期：

    2025年

查看全文
  • Ravuri Daniel,
  • Kilari Jyothsna Devi,
  • Bode Prasad,
  • B. Ratna Sunil,
  • G. Naga Deekshitha

摘要

This study explores the ResNet-50 model for classifying fundus camera images into severity stages of diabetic retinopathy (DR). This study resolves the issue of limited access to medical specialists in rural regions by improving the speed and accuracy of DR detection. DR was classified into five severity levels, including No DR, Mild, Moderate, Severe, and Proliferative DR, using a dataset containing 1928 test images and 3363 training images. Image augmentation techniques were applied to improve generalization and model performance. ResNet-50 outperformed other architectures, including DenseNet, VGG16, AlexNet, InceptionV3, and classical CNN, with a specificity of 94.8%; because of its efficient utilization of residual connections to overcome the gradient-vanishing venture, ResNet-50 permitted deep learning and up-to-the-minute feature extraction. The study successfully identified the DR data without compromising computational efficiency and showed decent reproducibility for DR identification. These results suggest that ResNet-50-based models are a potential approach for automated DR severity classification, which could improve early diagnosis and management of the disease in resource-poor settings.