Scholarlink

This paper presents a purely supervised solution for the DFUC2024 challenge without using the un-annotated data, focusing on the segmentation of diabetic foot ulcers (DFUs) using deep learning techniques. We compared two prominent architectures, UNet and DeepLabV3, evaluating their performance using a 5-fold cross-validation (CV) strategy. The DeepLabV3 model, leveraging a ResNet50 backbone pretrained on ImageNet, outperformed the UNet model, achieving an average Dice Coefficient Score (DSC) of 0.7909 across the CV sets. Despite these results, the model demonstrated overfitting, with a reduced DSC of 0.6751 on the challenge test set, highlighting the challenges of generalization in data-limited scenarios.

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 Supervised Segmentation Solution: Diabetic Foot Ulcers Challenge 2024

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xue Feng,
  • Gaofeng Huang

摘要

This paper presents a purely supervised solution for the DFUC2024 challenge without using the un-annotated data, focusing on the segmentation of diabetic foot ulcers (DFUs) using deep learning techniques. We compared two prominent architectures, UNet and DeepLabV3, evaluating their performance using a 5-fold cross-validation (CV) strategy. The DeepLabV3 model, leveraging a ResNet50 backbone pretrained on ImageNet, outperformed the UNet model, achieving an average Dice Coefficient Score (DSC) of 0.7909 across the CV sets. Despite these results, the model demonstrated overfitting, with a reduced DSC of 0.6751 on the challenge test set, highlighting the challenges of generalization in data-limited scenarios.