Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Limited Supervised Foot Ulcer Segmentation Using Cross-Domain Augmentation

Shang-Jui Kuo, Po-Han Huang, Chia-Ching Lin, Jeng-Lin Li, Ming-Ching Chang

TL;DR

This work addresses the challenge of segmenting foot ulcers with very limited labeled data by using cross-domain data augmentation. The authors propose TransMix, which combines Augmented Global Pre-training (AGP) on a large skin dataset with Localized CutMix Fine-tuning (LCF) to enrich both global and local wound variations. In experiments on the FUSeg dataset with as few as 40–81 annotated images, TransMix significantly improves Dice scores over strong baselines, demonstrating effective domain transfer and augmentation. The approach offers a practical path to robust foot ulcer segmentation in real-world clinical settings and suggests extensions to few-shot or zero-shot learning.

Abstract

Diabetic foot ulcers pose health risks, including higher morbidity, mortality, and amputation rates. Monitoring wound areas is crucial for proper care, but manual segmentation is subjective due to complex wound features and background variation. Expert annotations are costly and time-intensive, thus hampering large dataset creation. Existing segmentation models relying on extensive annotations are impractical in real-world scenarios with limited annotated data. In this paper, we propose a cross-domain augmentation method named TransMix that combines Augmented Global Pre-training AGP and Localized CutMix Fine-tuning LCF to enrich wound segmentation data for model learning. TransMix can effectively improve the foot ulcer segmentation model training by leveraging other dermatology datasets not on ulcer skins or wounds. AGP effectively increases the overall image variability, while LCF increases the diversity of wound regions. Experimental results show that TransMix increases the variability of wound regions and substantially improves the Dice score for models trained with only 40 annotated images under various proportions.

Improving Limited Supervised Foot Ulcer Segmentation Using Cross-Domain Augmentation

TL;DR

This work addresses the challenge of segmenting foot ulcers with very limited labeled data by using cross-domain data augmentation. The authors propose TransMix, which combines Augmented Global Pre-training (AGP) on a large skin dataset with Localized CutMix Fine-tuning (LCF) to enrich both global and local wound variations. In experiments on the FUSeg dataset with as few as 40–81 annotated images, TransMix significantly improves Dice scores over strong baselines, demonstrating effective domain transfer and augmentation. The approach offers a practical path to robust foot ulcer segmentation in real-world clinical settings and suggests extensions to few-shot or zero-shot learning.

Abstract

Diabetic foot ulcers pose health risks, including higher morbidity, mortality, and amputation rates. Monitoring wound areas is crucial for proper care, but manual segmentation is subjective due to complex wound features and background variation. Expert annotations are costly and time-intensive, thus hampering large dataset creation. Existing segmentation models relying on extensive annotations are impractical in real-world scenarios with limited annotated data. In this paper, we propose a cross-domain augmentation method named TransMix that combines Augmented Global Pre-training AGP and Localized CutMix Fine-tuning LCF to enrich wound segmentation data for model learning. TransMix can effectively improve the foot ulcer segmentation model training by leveraging other dermatology datasets not on ulcer skins or wounds. AGP effectively increases the overall image variability, while LCF increases the diversity of wound regions. Experimental results show that TransMix increases the variability of wound regions and substantially improves the Dice score for models trained with only 40 annotated images under various proportions.
Paper Structure (9 sections, 2 figures, 1 table)

This paper contains 9 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Cross-Domain Transfer Strategies
  4. Augmented Global Pre-training
  5. Localized CutMix Fine-tuning
  6. Experimental Results
  7. Results Under Limited Supervision
  8. Analysis on the Amount Training Data
  9. Conclusion

Figures (2)

  • Figure 1: Overview of the proposed TransMix framework.Augmented Global Pre-training ( AGP) enables initial model learning from other larger skin datasets. Localized CutMix Fine-tuning ( LCF) expands the diversity of wound patches via CutMix synthesis. Our approach can effectively overcome the issue of data scarcity during model training.
  • Figure 2: We evaluate the model fine-tuned using different numbers of FUSeg training samples, and report the Dice scores (%) on the FUSeg validation set.