SymmAdapt: Symmetrical Flow Matching for Source-Free Domain Adaptation in Medical Image Segmentation

Tal Grossman; Noa Cahan; Hayit Greenspan

cs.CVAug 23, 2026

SymmAdapt: Symmetrical Flow Matching for Source-Free Domain Adaptation in Medical Image Segmentation

Authors: Tal Grossman, Noa Cahan, Hayit Greenspan

Organizations: Faculty of Engineering, Tel Aviv University, Tel Aviv, Israel

Abstract

Domain shift across imaging modalities and acquisition sites remains a significant barrier to the clinical deployment of segmentation models. Source-free unsupervised domain adaptation (SFUDA) addresses this by adapting a pretrained model to an unlabeled target domain without requiring access to sensitive source data. We introduce a novel SFUDA framework built on Symmetrical Flow Matching, a unified generative model that segments an input image and synthesizes a source-like image from a mask within the same learned flow. By initializing inference from a domain-agnostic Gaussian origin, the model preserves structural consistency across domains and grounds predictions in learned anatomy rather than shifted texture statistics. Our pipeline leverages this symmetry to generate reliable pseudo-labels and corresponding source-like synthetic images from unlabeled target data, creating a generative replay buffer that anchors source knowledge during a generative self-training stage that fine-tunes on a joint set of real target and synthetic source-like images. We evaluate on abdominal multi-organ and cardiac segmentation, covering cross-modality MRI<->CT shifts, and multi-site prostate segmentation. Our approach outperforms SFUDA baselines and is competitive with conventional UDA methods.

Figures & tables

Panel of Figure 1 — Figure 1: Symmetrical Flow Matching for medical tasks. SymmFlow learns a joint vector field from Gaussian noise to images ( $X$ ) and masks ( $Y$ ), enabling conditional segmentation ( $Y|X$ ) and synthesis ( $X|Y$ ). (a) Abdomen and cardiac examples produced by our source-trained model. (b) t-SNE shows overlapping mask features but separated image features. (c) Synthetic replay bridges the separated image-feature clusters.

Figure 2: SymmAdapt pipeline. Phase I: SymmFlow is trained on labeled source data, learning a bidirectional flow between images and masks. Phase II: At adaptation time (no source data), target images are segmented and the resulting pseudo-masks are used to generate synthetic source-like images; a round-trip consistency filter retains only cycle-consistent pairs. Phase III: The model is fine-tuned on a dual stream of real target and synthetic source-like pairs, anchoring source semantics while adapting to target appearance.

Method (MRI $\rightarrow$ CT)	SFUDA	Dice (% $\uparrow$ )					ASSD (mm $\downarrow$ )
Method (MRI $\rightarrow$ CT)	SFUDA	Liver	R.kidney	L.kidney	Spleen	Average	Liver	R.kidney	L.kidney	Spleen	Average
Target Supervised	–	95.4	89.6	90.3	94.2	92.4	0.64	1.39	0.58	0.52	0.78
SIFA [ 7 ]	$\times$	88.0	83.3	80.9	82.6	83.7	1.2	1.0	1.5	1.6	1.3
Diffusion-based DA [ 11 ]	$\times$	89.0	85.6	85.6	85.8	86.5	1.5	1.3	1.2	1.2	1.3
C3R [ 9 ]	$\times$	91.9	84.3	82.6	85.0	86.0	0.93	0.98	1.10	1.39	1.10
DPL [ 8 ]	✓	39.0	35.3	58.6	42.4	43.9	6.36	13.09	7.94	10.10	9.38

Table 1: Quantitative evaluation results on the abdominal multi-organ datasets.

Figure 3: Qualitative results across three benchmarks (rows: abdomen MRI $\to$ CT, cardiac MRI $\to$ CT, prostate RUNMC $\to$ QUBIQ). SymmAdapt yields smoother boundaries and fewer false positives, closer to the ground truth. The first column shows a representative source-domain image.

Method (MRI $\rightarrow$ CT)	SFUDA	Dice (% $\uparrow$ )					ASSD (mm $\downarrow$ )
Method (MRI $\rightarrow$ CT)	SFUDA	AA	LAC	LVC	MYO	Average	AA	LAC	LVC	MYO	Average
Target Supervised	–	94.8	92.3	92.1	89.8	92.2	1.96	4.32	2.14	1.67	2.52
SIFA [ 7 ]	$\times$	81.3	79.5	73.8	61.6	74.1	7.9	6.2	5.5	8.5	7.0
C3R [ 9 ]	$\times$	90.2	90.2	86.6	77.5	86.1	4.72	2.81	2.58	3.07	3.30
GenericSSL [ 19 ]	$\times$	93.2	89.5	91.7	86.2	90.1	–	–	–	–	1.7
AdaMI [ 1 ]	✓	83.1	78.2	74.5	66.8	75.7	5.6	4.2	5.7	6.9	5.6

Table 2: Quantitative evaluation results on the MMWHS cardiac dataset.

Method	SFUDA	RUNMC $\rightarrow$ QUBIQ		RUNMC $\rightarrow$ BMC
Method	SFUDA	Dice (% $\uparrow$ )	ASSD (mm $\downarrow$ )	Dice (% $\uparrow$ )	ASSD (mm $\downarrow$ )
Target Supervised	–	96.5	0.09	89.2	0.77
DPL [ 8 ]	✓	75.5	8.33	–	–
UPL [ 21 ]	✓	73.6	16.55	–	–
ProtoContra [ 22 ]	✓	86.6	5.58	–	–
AdaMI [ 1 ]	✓	81.0	8.46	79.5	3.92

Table 3: Quantitative evaluation results on the prostate task.

Explore similar work

Tell2Adapt: A Unified Framework for Source Free Unsupervised Domain Adaptation via Vision Foundation Model
Mar 5, 2026Yulong Shi, Shijie Li, Ziyi Li +1Source-Free Domain Adaptation Medical Image Segmentation
College of Medicine and Biological Information Engineering, Northeastern University · Engineering Research Center of Medical Imaging and Intelligent Analysis, Ministry of Education · Key Laboratory of Medical Image Computing, Ministry of Education, Northeastern University
SHAPE: Structure-aware Hierarchical Unsupervised Domain Adaptation with Plausibility Evaluation for Medical Image Segmentation
Mar 23, 2026Linkuan Zhou, Yinghao Xia, Yufei Shen +6Unsupervised Domain Adaptation Medical Image Segmentation
School of Software, Northwestern Polytechnical University · School of Computer Science and Technology, Harbin Institute of Technology · School of Software Engineering, USTC +3
One Size does not Fit All: Heterogeneous Latent Space Alignment for Unsupervised Domain Adaptation
Jun 19, 2026Evangelia Koskinioti, Yi Shen, Georgios Stamou +1Unsupervised Domain Adaptation Domain Adaptation
Department of Electrical and Computer Engineering, Duke University, NC, USA · Department of Mechanical Engineering, Duke University, NC, USA · Department of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece.

Mar 5, 2026J/K move · Enter open · S save

Tell2Adapt: A Unified Framework for Source Free Unsupervised Domain Adaptation via Vision Foundation Model

Yulong Shi, Shijie Li, Ziyi Li, Lin Qi

College of Medicine and Biological Information Engineering, Northeastern University · Engineering Research Center of Medical Imaging and Intelligent Analysis, Ministry of Education · Key Laboratory of Medical Image Computing, Ministry of Education, Northeastern University

Source-Free Domain Adaptation Medical Image Segmentation Vision Foundation Model Adaptation

Open paper AskPDF arXiv

Source Free Unsupervised Domain Adaptation (SFUDA) is critical for deploying deep learning models across diverse clinical settings. However, existing methods are typically designed for low-gap, specific domain shifts and cannot generalize into a unified, multi-modalities, and multi-target framework, which presents a major barrier to real-world application. To overcome this issue, we introduce Tell2Adapt, a novel SFUDA framework that harnesses the vast, generalizable knowledge of the Vision Foundation Model (VFM). Our approach ensures high-fidelity VFM prompts through Context-Aware Prompts Regularization (CAPR), which robustly translates varied text prompts into canonical instructions. This enables the generation of high-quality pseudo-labels for efficiently adapting the lightweight student model to target domain. To guarantee clinical reliability, the framework incorporates Visual Plausibility Refinement (VPR), which leverages the VFM's anatomical knowledge to re-ground the adapted model's predictions in target image's low-level visual features, effectively removing noise and false positives. We conduct one of the most extensive SFUDA evaluations to date, validating our framework across 10 domain adaptation directions and 22 anatomical targets, including brain, cardiac, polyp, and abdominal targets. Our results demonstrate that Tell2Adapt consistently outperforms existing approaches, achieving SOTA for a unified SFUDA framework in medical image segmentation. Code are avaliable at https://github.com/derekshiii/Tell2Adapt.

Abstract

Explore similar work

Tell2Adapt: A Unified Framework for Source Free Unsupervised Domain Adaptation via Vision Foundation Model

SHAPE: Structure-aware Hierarchical Unsupervised Domain Adaptation with Plausibility Evaluation for Medical Image Segmentation

One Size does not Fit All: Heterogeneous Latent Space Alignment for Unsupervised Domain Adaptation