Scholarlink

<p>Diabetic Macular Edema (DME) is a major cause of vision impairment among individuals with diabetes, particularly in settings where access to advanced retinal imaging such as Optical Coherence Tomography (OCT) is limited. This study explores a cost-conscious, AI-assisted approach that integrates retinal fundus images with clinical parameters to estimate macular thickness-a key biomarker for DME and to support DME classification. A dataset of 552 fundus images with corresponding clinical records was collected from two hospitals, Amrita Institute of Medical Sciences, Kochi, and Sabitha Eye Care Hospital, Pathanamthitta. Macular regions were annotated using the CVAT tool and segmented with a U-Net model. Six convolutional neural network models combining image-derived and clinical features were developed and compared against classical machine-learning methods XGBoost, AdaBoost, and Random Forest. Among the evaluated models, EfficientNetB0 demonstrated moderately strong performance across both tasks. In macular thickness estimation, it achieved an external-test R<sup>2</sup> of 0.80 (95% CI 0.73–0.87). When the predicted thickness values were used alongside clinical parameters for DME classification, EfficientNetB0 obtained an accuracy of 0.80 [0.74–0.86], precision of 0.78 [0.71–0.85], recall of 0.81 [0.77–0.89], F1-score of 0.80 [0.74–0.86], and an AUC of 0.82 [0.78–0.90]. These findings indicate that the proposed framework serve as a potential adjunct to existing diagnostic pathways by offering supportive quantitative insights where OCT availability is limited. The work aligns with the Sustainable Development Goals (SDGs), particularly SDG 3 (Good Health and Well-Being), SDG 9 (Industry, Innovation, and Infrastructure), and SDG 10 (Reduced Inequalities) by demonstrating a feasible, technology-enabled method that enhance decision support in resource-constrained environments.</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

Sustainable retinal diagnostics using deep learning framework for macular thickness estimation and diabetic macular edema detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月19日

查看全文
  • C. A. Aparna,
  • B. R. Manju

摘要

Diabetic Macular Edema (DME) is a major cause of vision impairment among individuals with diabetes, particularly in settings where access to advanced retinal imaging such as Optical Coherence Tomography (OCT) is limited. This study explores a cost-conscious, AI-assisted approach that integrates retinal fundus images with clinical parameters to estimate macular thickness-a key biomarker for DME and to support DME classification. A dataset of 552 fundus images with corresponding clinical records was collected from two hospitals, Amrita Institute of Medical Sciences, Kochi, and Sabitha Eye Care Hospital, Pathanamthitta. Macular regions were annotated using the CVAT tool and segmented with a U-Net model. Six convolutional neural network models combining image-derived and clinical features were developed and compared against classical machine-learning methods XGBoost, AdaBoost, and Random Forest. Among the evaluated models, EfficientNetB0 demonstrated moderately strong performance across both tasks. In macular thickness estimation, it achieved an external-test R2 of 0.80 (95% CI 0.73–0.87). When the predicted thickness values were used alongside clinical parameters for DME classification, EfficientNetB0 obtained an accuracy of 0.80 [0.74–0.86], precision of 0.78 [0.71–0.85], recall of 0.81 [0.77–0.89], F1-score of 0.80 [0.74–0.86], and an AUC of 0.82 [0.78–0.90]. These findings indicate that the proposed framework serve as a potential adjunct to existing diagnostic pathways by offering supportive quantitative insights where OCT availability is limited. The work aligns with the Sustainable Development Goals (SDGs), particularly SDG 3 (Good Health and Well-Being), SDG 9 (Industry, Innovation, and Infrastructure), and SDG 10 (Reduced Inequalities) by demonstrating a feasible, technology-enabled method that enhance decision support in resource-constrained environments.