Organizations: Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen, China · The University of Hong Kong, Hong Kong, China · Jiaxing Research Institute, Southern University of Science and Technology, Jiaxing, China · Department of Biomedical Engineering, Southern University of Science and Technology, Shenzhen, China
Distribution shift in medical imaging remains a central bottleneck for the clinical translation of medical AI. Failure to address it can lead to severe performance degradation in unseen environments and exacerbate health inequities. Existing methods for domain adaptation are inherently limited by exhausting predefined possibilities through simulated shifts or pseudo-supervision. Such strategies struggle in the open-ended and unpredictable real world, where distribution shifts are effectively infinite. To address this challenge, we introduce an empirical law called Rank Stability of Positive Regions'', which states that the relative rank of predicted probabilities for positive voxels remains stable under distribution shift. Guided by this principle, we propose CRISP, a parameter-free and model-agnostic framework requiring no target-domain information. CRISP is the first framework to make segmentation based on rank rather than probabilities. CRISP simulates model behavior under distribution shift via latent feature perturbation, where voxel probability rankings exhibit two stable patterns: regions that consistently retain high probabilities (destined positives according to the principle) and those that remain low-probability (can be safely classified as negatives). Based on these patterns, we construct high-precision (HP) and high-recall (HR) priors and recursively refine them under perturbation. We then design an iterative training framework, making HP and HR progressively squeeze'' to the final segmentation. Extensive evaluations on multi-center cardiac MRI and CT-based lung vessel segmentation demonstrate CRISP's superior robustness, significantly outperforming state-of-the-art methods with striking HD95 reductions of up to 0.14 (7.0% improvement), 1.90 (13.1% improvement), and 8.39 (38.9% improvement) pixels across multi-center, demographic, and modality shifts, respectively.
Figures & tables
Figure 1: Overview of CRISP, a rank-based segmentation framework rather than probability-driven prediction. (a) Three representative clinical shifts: demographic, modality, and protocol variations. (b) Empirical Rank Stability: the relative rank of foreground regions remains stable despite probability drifts across domains. (c) Latent Feature Perturbation: injecting Gaussian noise into bottleneck features to simulate domain shift, yielding stochastic rank maps. Across perturbations, consistently top-graded voxels form stable high-probability ( MHP ) core, while persistently non-background voxels form the stable non-background region ( MHR ). (d) Recursive Self-Evolution: guided by these hints, successive model generations progressively squeeze the ambiguity region between MHP and MHR , progressively bringing the two boundaries closer until convergence forms the final segmentation.
Metrics
Method
Target domain B
Target domain C
Target domain D
LV
MYO
RV
LV
MYO
RV
LV
MYO
RV
Dice ↑ (%)
Baseline
85.22±13.27
75.91±11.22
78.25±18.15
83.32±11.86
74.54±9.15
78.30±14.84
88.36±8.22
75.76±5.44
83.57±10.21
Upper Bound
90.92±7.09
85.10±3.73
86.47±9.39
88.60±8.58
81.54±7.12
85.78±7.12
92.75±4.79
81.67±3.30
87.56±17.83
FedDG ⋆ [ 14 ]
86.97±12.18
79.25±5.88
85.05±9.72
87.17±7.99
76.96±7.65
83.65±10.20
89.91±5.86
76.83±4.26
86.89±6.71
DDG-Med ⋆ [ 13 ]
89.01±9.78
80.15±5.08
84.81±9.51
88.67±6.70
78.62±6.39
83.33±8.55
90.75±3.27
78.40±4.94
85.34±7.11
IPLC # [ 11 ]
88.50±6.23
83.27±3.70
83.40±9.01
88.62±9.32
78.82±7.40
84.08±14.39
89.91±5.63
79.24±3.64
86.28±6.99
Table 1: Quantitative comparison on M&Ms dataset. Dice score (%) ↑ and HD95 (pixel) ↓ . The best results among domain generalization methods are highlighted in bold. ⋆ = Methods that cannot access target domain. # = Methods that need target domain to update parameters(need more information, need retraining).
Metrics
Method
Scenario 1 (Demographic)
Scenario 2 (Modality)
Normal Cohort (S)
COVID Cohort (T)
Contrast-enhanced (S)
Non-contrast (T)
Dice ↑ (%)
Baseline
81.14±4.24
65.15±16.62
74.47±3.86
33.26±8.48
IPLC # [ 11 ]
81.14±4.24
59.92±6.56
74.47±3.86
41.33±11.27
TEGDA # [ 12 ]
81.14±4.24
66.65±11.46
74.47±3.86
36.30±6.99
Ours ⋆
83.73±5.47
70.61±13.27
78.23±2.94
59.46±7.03
HD95 ↓ (pixel)
Baseline
4.84±2.43
17.12±25.04
8.24±2.78
29.97±13.38
Table 2: Quantitative evaluation across different clinical scenarios. (S) : Source domain, (T) : Target domain. Dice score (%) ↑ and HD95 (pixel) ↓ . ⋆ = Methods that cannot access target domain. # = Methods that need target domain to update parameters(need more information, need retraining).
Figure 2: Qualitative segmentation results of different methods. The top and bottom rows present the segmentation performance under modality shift (Non-contrast) and demographic shift (COVID Cohort), respectively.
Semi-supervised domain adaptation leverages a few labeled and many unlabeled target samples, making it promising for addressing domain shifts in medical image analysis. However, existing methods struggle with severity classification due to unclear class boundaries. Severity classification involves naturally ordered class labels, complicating adaptation. We propose a novel method that aligns source and target domains using rank scores learned via ranking with class order. Specifically, Cross-Domain Ranking ranks sample pairs across domains, while Continuous Distribution Alignment aligns rank score distributions. Experiments on ulcerative colitis and diabetic retinopathy classification validate the effectiveness of our approach, demonstrating successful alignment of class-specific rank score distributions.
Shota Harada, Ryoma Bise, Kiyohito Tanaka +1
Kyushu University, Japan · Kyoto Second Red Cross Hospital, Japan
We present MaskGen, a theoretically grounded and deliberately simple approach for domain generalization in 3D biomedical image segmentation. Modern segmentation models degrade sharply under shifts in modality, disease severity, clinical sites, and more, limiting their reliable adoption. Existing generalization methods address this using extreme augmentations, hand-engineered domain statistics mixing, or architectural redesigns that add significant implementation overhead while yielding inconsistent performance across biomedical settings. MaskGen instead presents a principled learning strategy with marginal overhead that utilizes both source-domain image intensities and domain-stable foundation model representations to train robust segmentation models. As a result, MaskGen achieves strong gains in both fully supervised and few-shot segmentation across broad clinical shifts in biomedical studies. Unlike prior approaches, MaskGen is architecture- and loss-agnostic, compatible with standard augmentation pipelines, easy to implement, and tackles arbitrary anatomical regions. Its implementation is freely available at https://github.com/sebodiaz/MaskGen.
Unsupervised Domain Adaptation (UDA) is essential for deploying medical segmentation models across diverse clinical environments. Existing methods are fundamentally limited, suffering from semantically unaware feature alignment that results in poor distributional fidelity and from pseudo-label validation that disregards global anatomical constraints, thus failing to prevent the formation of globally implausible structures. To address these issues, we propose SHAPE (Structure-aware Hierarchical Unsupervised Domain Adaptation with Plausibility Evaluation), a framework that reframes adaptation towards global anatomical plausibility. Built on a DINOv3 foundation, its Hierarchical Feature Modulation (HFM) module first generates features with both high fidelity and class-awareness. This shifts the core challenge to robustly validating pseudo-labels. To augment conventional pixel-level validation, we introduce Hypergraph Plausibility Estimation (HPE), which leverages hypergraphs to assess the global anatomical plausibility that standard graphs cannot capture. This is complemented by Structural Anomaly Pruning (SAP) to purge remaining artifacts via cross-view stability. SHAPE significantly outperforms prior methods on cardiac and abdominal cross-modality benchmarks, achieving state-of-the-art average Dice scores of 90.08% (MRI->CT) and 78.51% (CT->MRI) on cardiac data, and 87.48% (MRI->CT) and 86.89% (CT->MRI) on abdominal data. The code is available at https://github.com/BioMedIA-repo/SHAPE.
Linkuan Zhou, Yinghao Xia, Yufei Shen +6
School of Software, Northwestern Polytechnical University · School of Computer Science and Technology, Harbin Institute of Technology · School of Software Engineering, USTC +3