Scholarlink

Background <p>Macular edema is a vision-threatening complication of diabetic retinopathy that requires early detection and intervention to prevent severe visual impairment. Conventional diagnosis relies heavily on the expertise of ophthalmologists, which may be resource-intensive and time-consuming, particularly in large-scale screening scenarios. Artificial intelligence (AI) offers an opportunity to enhance efficiency and accuracy in such diagnostic processes.</p> Objective <p>This study aims to develop and evaluate an Enhanced Multi-feature Fusion Network (MFFN) for the automated detection and grading of diabetic macular edema (DME) from fundus images, supporting ophthalmologists in timely and accurate decision-making.</p> Methods <p>We utilized fundus images from diabetic retinopathy patients as the primary dataset to design a deep learning-based MFFN model for detecting hard exudates and classifying DME severity. The model integrates multi-feature fusion to capture spatial and contextual cues, while Grad-CAM visualization was employed to enhance interpretability and transparency in AI predictions. Performance was evaluated on benchmark datasets and in real-world clinical settings.</p> Results <p>The proposed MFFN achieved an accuracy of 99.8% for DME detection and 99.5% for severity grading on benchmark datasets. In real-world clinical validation, the accuracy was 97.9%, reflecting high performance but also revealing potential challenges in generalizing to diverse demographic and imaging conditions. Grad-CAM visualizations provided clear localization of pathological regions, aiding interpretability and clinical trust.</p> Conclusions <p>The MFFN demonstrates strong potential for cost-effective, rapid, and large-scale screening of macular edema, offering a valuable AI-assisted diagnostic tool for ophthalmologists. While preliminary results are highly promising, further validation on larger and more diverse patient populations is essential to ensure robustness and clinical applicability. Future work will focus on optimizing generalizability and integrating the model into practical healthcare workflows.</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

Advanced fusion network for detecting and grading macular edema in diabetic patients

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • C. Kotteeswari,
  • S. N. Sangeethaa,
  • S. Jothimani

摘要

Background

Macular edema is a vision-threatening complication of diabetic retinopathy that requires early detection and intervention to prevent severe visual impairment. Conventional diagnosis relies heavily on the expertise of ophthalmologists, which may be resource-intensive and time-consuming, particularly in large-scale screening scenarios. Artificial intelligence (AI) offers an opportunity to enhance efficiency and accuracy in such diagnostic processes.

Objective

This study aims to develop and evaluate an Enhanced Multi-feature Fusion Network (MFFN) for the automated detection and grading of diabetic macular edema (DME) from fundus images, supporting ophthalmologists in timely and accurate decision-making.

Methods

We utilized fundus images from diabetic retinopathy patients as the primary dataset to design a deep learning-based MFFN model for detecting hard exudates and classifying DME severity. The model integrates multi-feature fusion to capture spatial and contextual cues, while Grad-CAM visualization was employed to enhance interpretability and transparency in AI predictions. Performance was evaluated on benchmark datasets and in real-world clinical settings.

Results

The proposed MFFN achieved an accuracy of 99.8% for DME detection and 99.5% for severity grading on benchmark datasets. In real-world clinical validation, the accuracy was 97.9%, reflecting high performance but also revealing potential challenges in generalizing to diverse demographic and imaging conditions. Grad-CAM visualizations provided clear localization of pathological regions, aiding interpretability and clinical trust.

Conclusions

The MFFN demonstrates strong potential for cost-effective, rapid, and large-scale screening of macular edema, offering a valuable AI-assisted diagnostic tool for ophthalmologists. While preliminary results are highly promising, further validation on larger and more diverse patient populations is essential to ensure robustness and clinical applicability. Future work will focus on optimizing generalizability and integrating the model into practical healthcare workflows.