Scholarlink

Center-involved diabetic macular edema (ci-DME) is a sight-threatening complication of diabetes. Early detection is crucial for preventing vision loss. This paper proposes a deep learning-based method using Ultra-Wide Field Color Fundus Photography (UWF-CFP) images to predict ci-DME development within one year. The challenge lies in handling images from different devices (Optos, Clarus, Eidon) without access to training data. Our approach employs a two-step model: first, a device detection model identifies the imaging device, followed by device-specific classification models for ci-DME prediction. The proposed method achieves an AUC score of 0.9860 on a simulated validation set. However, further refinement is needed, as evidenced by the low F1 score and calibration error. We discuss potential improvements using transformers and foundation models like RetFound for enhanced ci-DME prediction.

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

A Device-Agnostic Deep Learning Approach for Predicting Ci-DME Onset Using UWF-CFP Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Amin Khouani,
  • Ihsane Mekki

摘要

Center-involved diabetic macular edema (ci-DME) is a sight-threatening complication of diabetes. Early detection is crucial for preventing vision loss. This paper proposes a deep learning-based method using Ultra-Wide Field Color Fundus Photography (UWF-CFP) images to predict ci-DME development within one year. The challenge lies in handling images from different devices (Optos, Clarus, Eidon) without access to training data. Our approach employs a two-step model: first, a device detection model identifies the imaging device, followed by device-specific classification models for ci-DME prediction. The proposed method achieves an AUC score of 0.9860 on a simulated validation set. However, further refinement is needed, as evidenced by the low F1 score and calibration error. We discuss potential improvements using transformers and foundation models like RetFound for enhanced ci-DME prediction.