Scholarlink

Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients. Early identification of DR relies highly on standard color fundus photography (CFP), which faces limitations due to a restricted field of view and suboptimal image quality. Ultra-widefield (UWF) fundus images, offering broader retinal coverage, have emerged as a promising alternative. However, diagnosing DR based on UWF images is often time-consuming and laborious, highlighting the need for a unified, automated solution. In this work, we present a highly optimized lightweight Convolutional Neural Networks (CNNs) training pipeline based on MobileNet for the automated classification of diabetic macular edema (DME), a significant complication of DR. Our model significantly accelerates the analysis of UWF images compared to manual methods, enabling faster DR diagnosis and improved patient management. Notably, our approach achieved third place in the MICCAI UWF4DR 2024 Challenge for the Classification of Diabetic Macular Edema. The pipeline is publicly available at https://github.com/Retinal-Research/NN-MOBILENET .

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

DME-MobileNet: Fine-Tuning nnMobileNet for Diabetic Macular Edema Classification

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xuanzhao Dong,
  • Yalin Wang,
  • Yanxi Chen,
  • Xin Li,
  • Hao Wang,
  • Peijie Qiu,
  • Xiwen Chen,
  • Abolfazl Razi,
  • Yujian Xiong,
  • Oana M. Dumitrascu,
  • Wenhui Zhu

摘要

Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients. Early identification of DR relies highly on standard color fundus photography (CFP), which faces limitations due to a restricted field of view and suboptimal image quality. Ultra-widefield (UWF) fundus images, offering broader retinal coverage, have emerged as a promising alternative. However, diagnosing DR based on UWF images is often time-consuming and laborious, highlighting the need for a unified, automated solution. In this work, we present a highly optimized lightweight Convolutional Neural Networks (CNNs) training pipeline based on MobileNet for the automated classification of diabetic macular edema (DME), a significant complication of DR. Our model significantly accelerates the analysis of UWF images compared to manual methods, enabling faster DR diagnosis and improved patient management. Notably, our approach achieved third place in the MICCAI UWF4DR 2024 Challenge for the Classification of Diabetic Macular Edema. The pipeline is publicly available at https://github.com/Retinal-Research/NN-MOBILENET .