Scholarlink

In this paper, we present a comprehensive summary of our proposed methodology and experimental results for Task3 in the Ultra-Widefield Fundus (UWF) Imaging for Diabetic Retinopathy Challenge 2024, which focuses on the precise identification of Diabetic Macular Edema (DME). The primary challenges of Task3 lie in two aspects: firstly, the limited availability of training samples poses a risk of overfitting deep learning-based model; secondly, the UWF images encompass a wider retinal area, thereby augmenting complexity associated with image interpretation. To address these challenges, we have devised specialized training strategies and developed a novel model based on Swin-Transformer v2 to achieve accurate DME identification. Our approach has demonstrated exceptional performance with an AUROC value of 0.9835 during the testing phase. The code is available at: https://github.com/yourfla/UWF-DME-AutoID .

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

Automatic Identification Method for Diabetic Macular Edema in Ultra-widefield Fundus Images

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Heyou Chang,
  • Zhikang Ge,
  • Jian Zhang,
  • Heng Zhang,
  • Hao Zheng

摘要

In this paper, we present a comprehensive summary of our proposed methodology and experimental results for Task3 in the Ultra-Widefield Fundus (UWF) Imaging for Diabetic Retinopathy Challenge 2024, which focuses on the precise identification of Diabetic Macular Edema (DME). The primary challenges of Task3 lie in two aspects: firstly, the limited availability of training samples poses a risk of overfitting deep learning-based model; secondly, the UWF images encompass a wider retinal area, thereby augmenting complexity associated with image interpretation. To address these challenges, we have devised specialized training strategies and developed a novel model based on Swin-Transformer v2 to achieve accurate DME identification. Our approach has demonstrated exceptional performance with an AUROC value of 0.9835 during the testing phase. The code is available at: https://github.com/yourfla/UWF-DME-AutoID .