Scholarlink

In this work, we presented a novel DR diagnosis model utilizing graph neural networks (GNN) & transferring learning over retinal fundus images. To portray the relationships between the various retinal regions, we tested a graph using CNN as a feature extractor. A GNN enhances diagnostic precision and interpretability by classifying images into several DR phases. Our approach surpasses the most advanced CNN-based techniques when tested on Kaggle’s Diabetic Retinopathy Detection, Messidor-2, and APTOS 2019 datasets, obtaining 98.2% accuracy with 97.8% sensitivity and 98.5% specificity. Our method’s result is shown by the statistical significance test (p < 0.001).

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

Diagnosis of Diabetic Retinopathy Using Transfer Learning and Graph Neural Networks with Fundus Imaging

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sailakshmi Lakkakula,
  • Jonnadula Narasimharao,
  • Tarak Hussain

摘要

In this work, we presented a novel DR diagnosis model utilizing graph neural networks (GNN) & transferring learning over retinal fundus images. To portray the relationships between the various retinal regions, we tested a graph using CNN as a feature extractor. A GNN enhances diagnostic precision and interpretability by classifying images into several DR phases. Our approach surpasses the most advanced CNN-based techniques when tested on Kaggle’s Diabetic Retinopathy Detection, Messidor-2, and APTOS 2019 datasets, obtaining 98.2% accuracy with 97.8% sensitivity and 98.5% specificity. Our method’s result is shown by the statistical significance test (p < 0.001).