Semi-Supervised Learning with Foundation Models for Biomedical Data Analysis: Multimodality, Generative AI, and Clinical Adaptation

Biomedical data analysis has evolved rapidly from convolutional neural network-based systems toward transformer architectures and large-scale foundation models. At the same time, semi-supervised learning has progressed from classic pseudo-labeling and consistency regularization to more advanced approaches that leverage self-supervised learning, cross-modal representation alignment, and multimodal reasoning. These developments are especially important in medicine, where imaging, clinical text, and other heterogeneous data sources are increasingly available, but expert annotation remains scarce. The convergence of semi-supervised learning, vision-language modeling, and generative AI creates new opportunities for learning from unlabeled or weakly labeled biomedical data. However, important challenges remain, including reliable adaptation to clinical settings, robustness to domain shifts, factual correctness in generative systems, and the need for clinically meaningful explanations.This Research Topic aims to address a central challenge in biomedical artificial intelligence: how to effectively adapt powerful foundation models to clinical tasks when high-quality annotations are limited. Although recent advances in vision transformers, segment-anything-style models, and large language models have significantly improved biomedical data analysis, their success often depends on large fully labeled datasets that are difficult and expensive to obtain in healthcare settings. We therefore seek contributions that explore semi-supervised learning as a practical and methodological bridge between large-scale pretraining and real-world clinical deployment. The goal is to bring together researchers working on data-efficient learning, multimodal modeling, and clinically grounded AI, and to promote methods that can better exploit unlabeled biomedical data while improving robustness, interpretability, and translational value in medical applications. We welcome original research articles, reviews, and perspective papers that advance the theory or application of semi-supervised learning in biomedical data analysis. Topics of interest include, but are not limited to:- Methodological innovations in semi-supervised learning- Semi-supervised disease classification, detection, and segmentation across imaging and multimodal data- Adaptation of foundation models to biomedical tasks- Integration of large language models and vision-language models into semi-supervised pipelines- Weak supervision and hybrid learning strategies- Multimodal representation learning- Federated and privacy-preserving semi-supervised learning- Test-time adaptation- Explainability and reliability- Robustness under data shiftWe particularly encourage submissions that demonstrate methodological novelty, strong clinical relevance, and clear translational value, rather than studies that only apply existing pretrained models to new datasets without substantive technical innovation. Contributions should also clearly discuss how their methods support clinical decision-making and practical deployment in resource-constrained healthcare environments. Topic editors: Qiangguo Jin, Leyi Wei, Ran Su Source: https://www.frontiersin.org/research-topics/79902/semi-supervised-learning-with-foundation-models-for-biomedical-data-analysis-multimodality-generative-ai-and-clinical-adaptation