Scholarlink

<p>Glaucoma is a serious eye disease that causes permanent blindness if not detected early. Since the disease develops slowly and the changes in the optic nerve are very subtle, early diagnosis from retinal fundus images is challenging. In this paper, we propose a new method called Correlation-guided Meta-Fusion based Dual Network (CoMFDNet) for automatic glaucoma classification. Our model uses two feature extractors: ResNet50, which captures deep structural and semantic features of the optic nerve head, and MobileNetV2, which provides lightweight texture-level features for faster processing. To combine these features, we have used a correlation-based fusion module that measures similarity between features using cosine similarity. This helps the model keep useful and complementary features while reducing redundant information. We also use Efficient Channel Attention (ECA) and Spatial Attention to highlight the most important retinal regions, such as the optic disc and cup, which are key in glaucoma diagnosis. Finally, a meta-learner adaptively adjusts the contribution of each branch, making the model more flexible and robust. We have tested our model on two standard glaucoma datasets. On the ACRIMA dataset, CoMFDNet achieves an accuracy of 99.63% and an F1-score of 99.62%. On the LAG dataset, it achieves an accuracy of 97.32% and an F1-score of 97.08%. The results show that our approach outperforms existing methods while remaining efficient for real-world use in clinical screening. The code implementation of the methodology is available at: <a href="https://github.com/Cmatermedicalimageanalysis/CoMFDNet.git">https://github.com/Cmatermedicalimageanalysis/CoMFDNet.git</a></p>

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

CoMFDNet: a correlation-guided meta-fusion based dual network for glaucoma classification

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Ananya Saha,
  • Gouranga Maity,
  • Ram Sarkar

摘要

Glaucoma is a serious eye disease that causes permanent blindness if not detected early. Since the disease develops slowly and the changes in the optic nerve are very subtle, early diagnosis from retinal fundus images is challenging. In this paper, we propose a new method called Correlation-guided Meta-Fusion based Dual Network (CoMFDNet) for automatic glaucoma classification. Our model uses two feature extractors: ResNet50, which captures deep structural and semantic features of the optic nerve head, and MobileNetV2, which provides lightweight texture-level features for faster processing. To combine these features, we have used a correlation-based fusion module that measures similarity between features using cosine similarity. This helps the model keep useful and complementary features while reducing redundant information. We also use Efficient Channel Attention (ECA) and Spatial Attention to highlight the most important retinal regions, such as the optic disc and cup, which are key in glaucoma diagnosis. Finally, a meta-learner adaptively adjusts the contribution of each branch, making the model more flexible and robust. We have tested our model on two standard glaucoma datasets. On the ACRIMA dataset, CoMFDNet achieves an accuracy of 99.63% and an F1-score of 99.62%. On the LAG dataset, it achieves an accuracy of 97.32% and an F1-score of 97.08%. The results show that our approach outperforms existing methods while remaining efficient for real-world use in clinical screening. The code implementation of the methodology is available at: https://github.com/Cmatermedicalimageanalysis/CoMFDNet.git