HEAL: Learning-Free Source Free Unsupervised Domain Adaptation for Cross-Modality Medical Image Segmentation

Yulong Shi, Jiapeng Li, Lin Qi.
The 36th British Machine Vision Conference (BMVC), 2025. CCF-C, CORE A

Abstract

HEAL introduces a learning-free approach to source-free unsupervised domain adaptation for cross-modality medical image segmentation. The method eliminates the need for source data access during adaptation, providing an efficient solution for deploying segmentation models across different medical imaging modalities.

Method Overview

The framework achieves domain adaptation without any additional training or fine-tuning. By leveraging intrinsic properties of medical images and domain-specific priors, HEAL enables direct deployment of pre-trained segmentation models to new imaging modalities while maintaining competitive performance.

Resources

• Paper (PDF)
• arXiv
• Code (GitHub)

BibTeX

@misc{shi2025heallearningfreesourcefree,
      title={HEAL: Learning-Free Source Free Unsupervised Domain Adaptation for Cross-Modality Medical Image Segmentation},
      author={Yulong Shi and Jiapeng Li and Lin Qi},
      year={2025},
      eprint={2511.17958},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.17958},
}