Scholarlink

Diabetic retinopathy (DR) is a major complication of diabetes that may lead to vision loss and other cascading complications. This research focuses on unifying the prediction capabilities of three popular deep convolutional neural networks—EfficientNet-B0, ResNet-50, and DenseNet-121 in an ensemble framework when trained on fundus images of the retina. A soft voting mechanism is employed to fuse the posterior class probabilities predicted by the three models. The high classification accuracy (84.59%), F1-score (0.8053), and G-mean score (0.5275) achieved on a benchmark DR detection dataset proves the efficacy of the soft voting ensemble classifier. The results establish the potential of the proposed soft voting ensemble approach in influencing future research on integrated deep learning frameworks for DR diagnosis.

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

Diabetic Retinopathy Detection Using Deep Learning Ensemble with Soft Voting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Akshat Barwal,
  • Akshat Rohil,
  • Aman Goyal,
  • Seba Susan

摘要

Diabetic retinopathy (DR) is a major complication of diabetes that may lead to vision loss and other cascading complications. This research focuses on unifying the prediction capabilities of three popular deep convolutional neural networks—EfficientNet-B0, ResNet-50, and DenseNet-121 in an ensemble framework when trained on fundus images of the retina. A soft voting mechanism is employed to fuse the posterior class probabilities predicted by the three models. The high classification accuracy (84.59%), F1-score (0.8053), and G-mean score (0.5275) achieved on a benchmark DR detection dataset proves the efficacy of the soft voting ensemble classifier. The results establish the potential of the proposed soft voting ensemble approach in influencing future research on integrated deep learning frameworks for DR diagnosis.