HEAL: Learning-Free Source Free Unsupervised Domain Adaptation for Cross-Modality Medical Image Segmentation
Yulong Shi, Jiapeng Li, Lin Qi
TL;DR
SFUDA aims to bridge domain gaps in medical image segmentation without accessing source data or target labels. HEAL achieves learning-free adaptation at inference by using a diffusion-conditioned, source-like sample generation, followed by Hierarchical Denoising ($Y_T^{entropy}$ and $Y_T^*$), Edge-Guided Selection to pick structurally consistent samples, and Size-Aware Fusion to combine predictions into $\hat{Y}_T$. Across BraTS 2021 and polyp datasets, HEAL sets new state-of-the-art results in cross-modality directions, outperforming No Adaptation and existing SFUDA methods. The approach offers strong privacy preservation and computational efficiency by avoiding target-domain training while delivering robust segmentation performance.
Abstract
Growing demands for clinical data privacy and storage constraints have spurred advances in Source Free Unsupervised Domain Adaptation (SFUDA). SFUDA addresses the domain shift by adapting models from the source domain to the unseen target domain without accessing source data, even when target-domain labels are unavailable. However, SFUDA faces significant challenges: the absence of source domain data and label supervision in the target domain due to source free and unsupervised settings. To address these issues, we propose HEAL, a novel SFUDA framework that integrates Hierarchical denoising, Edge-guided selection, size-Aware fusion, and Learning-free characteristic. Large-scale cross-modality experiments demonstrate that our method outperforms existing SFUDA approaches, achieving state-of-the-art (SOTA) performance. The source code is publicly available at: https://github.com/derekshiii/HEAL.