Scholarlink

This study presents an evaluation of three deep learning approaches submitted to the MICCAI’s 2024 DIAMOND (Device-Independent diAbetic Macular edema ONset preDiction) challenge, aimed at predicting the onset of center-involved diabetic macular edema (ci-DME) within a year based on Ultra-Widefield Fundus Images. Participants didn’t have access to the training, validation, or test datasets, making it necessary to develop models solely using a DeepDRiD synthetic dataset. We experimented with various deep learning architectures, including ResNet101, InceptionV3, and ResNet152, alongside different loss functions such as cross-entropy and class-balanced focal loss, complemented by data augmentation techniques. Our best-performing submission, using ResNet152, achieved an AUC of 0.798, an F1 score of 0.65, and an ECE of 0.11 on the DeepDRiD synthetic dataset.

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

Onset Prediction of Center-Involved Diabetic Macular Edema on Ultra-Widefield Fundus Images Using Finetuned ResNet with Class Balanced Focal Loss

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yasine Lehmiani,
  • Ayoub Elkhouzari,
  • Abdelhak Mahmoudi

摘要

This study presents an evaluation of three deep learning approaches submitted to the MICCAI’s 2024 DIAMOND (Device-Independent diAbetic Macular edema ONset preDiction) challenge, aimed at predicting the onset of center-involved diabetic macular edema (ci-DME) within a year based on Ultra-Widefield Fundus Images. Participants didn’t have access to the training, validation, or test datasets, making it necessary to develop models solely using a DeepDRiD synthetic dataset. We experimented with various deep learning architectures, including ResNet101, InceptionV3, and ResNet152, alongside different loss functions such as cross-entropy and class-balanced focal loss, complemented by data augmentation techniques. Our best-performing submission, using ResNet152, achieved an AUC of 0.798, an F1 score of 0.65, and an ECE of 0.11 on the DeepDRiD synthetic dataset.