This work proposes a label-free selection criterion built on SUDO, a framework for evaluating clinical AI systems without ground-truth annotations, and shows that AURCC can be used to rank a variety of vision-language models on chest X-ray classification across three inter-hospital shift scenarios, under zero-shot and MLP-probe regimes.
Abstract
Foundation models are increasingly deployed for medical image analysis. However, under the inter-institutional distribution shift typical of deployment, their performance varies widely and cannot be known without target-domain labels, which are rarely available. This leaves a practical question unresolved: given several candidate foundational models and labeled-data from a source domain, which one to deploy in an unlabeled target domain? We propose a label-free selection criterion built on SUDO, a framework for evaluating clinical AI systems without ground-truth annotations. SUDO partitions the unlabeled target data by predicted probability and, for each region, measures a pseudo-label discrepancy reflecting class contamination; aggregated across regions, this yields a score (AURCC) requiring neither target annotation nor fine-tuning. We show that AURCC can be used to rank a variety of vision-language models (BioMedCLIP, CXR-CLIP, CheXzero, MedCLIP, MedImageInsight, CLIP) on chest X-ray classification across three inter-hospital shift scenarios, under zero-shot and MLP-probe regimes. The AURCC ranking recovers the ground-truth ranking with Spearman rho up to 0.943 (p<0.05). Against the natural baseline of ranking by held-out source accuracy, AURCC is competitive when the labeled source is large and yields a more accurate ranking once it is small; the regime of interest in resource-constrained settings.
Prelocalized focal liver lesion classification from multi-phase magnetic resonance imaging remains challenging in multi-center clinical practice because lesion-level annotation is limited, imaging protocols vary across institutions, and routine unlabeled data may contain categories outside the predefined training taxonomy. Existing semi-supervised methods usually assume a closed label space and can therefore be affected by unreliable pseudo-labels when unknown or weakly supported lesions appear in the unlabeled pool. These challenges are closely related to out-of-distribution (OOD) effects caused by category mismatch and clinical distribution shift. To address this problem, we propose OOD-aware Reliability Learning (ORL), a semi-supervised framework for prelocalized liver lesion classification under sparse annotation, partial label-space mismatch, and multi-center distribution shift. ORL learns a compact known-class reference manifold using Prototype-Constrained Representation Learning (PCRL), estimates sample-wise manifold support through Transport-derived Compatibility Estimation (TCE) with Asymmetric Relaxed Optimal Transport (AROT), and combines this support with classifier confidence and prototype affinity to regulate pseudo-label learning. The framework also learns an amortized reliability predictor for inference-time support scoring and reliability-based case ranking after lesion localization. We evaluated ORL on a multi-center liver magnetic resonance imaging cohort and selected retrospective stress settings, including external-center evaluation, unlabeled-pool contamination, missing-phase testing, and composite OOD-oriented score analyses. ORL improved known-category classification over representative semi-supervised baselines and maintained better performance under contamination and distribution shift. These results indicate that manifold support estimation can improve label-efficient prelocalized liver lesion classification and may improve reliability-based case ranking after lesion localization in the evaluated retrospective setting.
Yuling Pu, Wei Xia, Lin Deng et al.· Computerized Medical Imaging...· 0 citations
Pretrained image encoders are central to medical image classification, where expert annotation is costly and task-specific cohorts are often limited. As the model space expands from general-purpose to broad-medical and specialty-specific encoders, selecting the representation becomes a substantive modeling decision. Clean-test discrimination alone is insufficient for this purpose: encoders with similar AUROC can differ in calibration, label efficiency, and stability under acquisition perturbations or distribution shift. We introduce CRS-Bench, a controlled benchmark for multi-objective medical encoder selection. CRS-Bench evaluates 15 pretrained encoder families across dermatology, ophthalmology, and radiology using ISIC 2019, APTOS 2019, and CheXpert, with CheXpert-to-MIMIC-CXR as an observed institutional shift, yielding 17,575 controlled run records and 3,515 seed-aggregated metric rows. Each encoder is characterized along four operational reliability dimensions: discrimination, calibration, label efficiency, and robustness. We summarize these dimensions using the Clinical Reliability Score (CRS), a Pareto-aware, reference-relative score combining dominance, profile balance, and worst-axis performance. AUROC and CRS are positively associated but not decision-equivalent: 21 of 105 pairwise orderings reverse, with a mean absolute rank displacement of 1.87. Paired-seed bootstrap analysis identifies PanDerm, MedSigLIP, and MedGemma as a stable leading reliability tier rather than a statistically resolved single leader. CRS-Bench provides a controlled framework for selecting medical image encoders from multi-axis reliability profiles rather than clean-test AUROC alone.
Xingtao Lin, Hangqi Ren, Caiwan Sun et al.· 0 citations
Deploying unsupervised domain adaptation (UDA) in clinical practice requires choosing which algorithm to use and which of its trained models to ship. However, the deployment (target) domain is unlabeled, so models cannot be evaluated directly on it, leaving it unclear which to select. We address this by evaluating the complete UDA pipeline, considering both adaptation and label-free selection together. Our study covers eleven clinically relevant cross-domain scenarios from nine medical imaging datasets, with ten UDA algorithms and 13 label-free selection methods (validators), evaluating over 80,000 trained models in total. By this, we find that a capable adapted model usually exists, but identifying it without target labels is difficult: the validator-selected models leave a large and structural target performance gap to the best available one, with no evaluated validator consistently reliable. Towards closing it, we explore two strategies, ensembling and a small target-labeling budget; both narrow this gap but do not close it entirely. Overall, deployable UDA depends on the complete pipeline; addressing the less explored selection step could bring much of current UDA closer to clinical use.
Yi Xiong, Luisa Gallée, D. Wolf et al.· 1 citation
Medical image segmentation is moving from task-specific convolutional models toward foundation models that can be adapted across organs, modalities, and clinical tasks with fewer manual labels. This transition has been accelerated by self-supervised pretraining, vision-language learning, and promptable segmentation frameworks such as the Segment Anything Model and its medical derivatives. However, the clinical value of these systems cannot be inferred from technical novelty alone. Medical images differ from natural images in dimensionality, intensity statistics, acquisition protocols, disease prevalence, and safety requirements, and recent evaluations show that naive zero-shot transfer remains inconsistent across modalities and lesion types. This narrative review synthesizes literature published up to May 22, 2026, on foundation models for medical image segmentation, with emphasis on technical evolution, application scenarios, validation strategies, and governance needs. Current evidence suggests that foundation models are most promising when they are deployed as interactive, auditable components within human-in-the-loop workflows, where they can reduce annotation burden, support rapid draft segmentation, and improve consistency across large imaging studies. Their translation into routine practice requires external validation, uncertainty-aware quality control, prospective workflow evaluation, bias assessment, and transparent reporting under medical AI guidelines. Future work should prioritize patient-level multimodal modeling, 3D and longitudinal segmentation, federated evaluation, and clinically meaningful endpoints rather than isolated benchmark gains.
Juntao Wei· Scientific Journal of Techno...· 0 citations
Numerous unsupervised domain adaptation (UDA) algorithms exist, but for clinical practice, selecting the best-suited one along with proper hyperparameters often remains unclear, as the unlabeled deployment (target) domain prevents direct evaluation. We propose a label-free criterion that jointly selects the algorithm and hyperparameters for UDA. Given a pool of candidate models from multiple algorithms trained with different hyperparameters, our approach scores each candidate against an agreement reference, and selects the one with the highest score. The agreement reference is constructed in two levels without using target labels. First, we leverage multiple label-free selection signals, using each to nominate a model within every algorithm. Second, the nominated models are aggregated across algorithms to form a reference prediction for each unlabeled target sample. The candidate whose predictions agree most with this reference is then selected for deployment. Experimental results on four brain MRI and four chest X-ray datasets across seven clinically relevant transfer scenarios show that our method achieves better selection performance than other methods and remains effective across different algorithm pools. Our approach takes a step towards practical, label-free algorithm selection for clinical deployment of UDA.
Medical imaging models often degrade when deployed at new clinical sites due to differences in imaging equipment, protocols, and patient populations. Test-time adaptation (TTA) addresses this by updating a pretrained model using only unlabeled target data, without access to source data. However, existing TTA methods were designed for single-label classification on natural image benchmarks, minimizing entropy uniformly across all samples without considering label dependencies. This overlooks a key property of multi-label medical imaging: pathologies do not occur independently but exhibit structured co-occurrence patterns. In this work, we propose Co-occurrence Weighted Adaptation (CoWA), which leverages disease co-occurrence patterns as a reliability signal for adaptation. CoWA estimates label co-occurrence structure from model predictions and downweights samples that deviate from expected patterns, enabling adaptation to rely more on consistent predictions while reducing the impact of noisy ones. We evaluate CoWA on chest X-ray benchmarks under domain shifts and demonstrate consistent improvements over established baselines.
Woo-Kyoung Jeong, Yujin Choi, Dongbin Kim et al.· 0 citations