Foundation models in biomedical imaging: turning hype into reality

Abstract

Foundation models (FMs) are driving a prominent shift in biomedical imaging, from task-specific models to unified backbone models for diverse tasks. This opens an avenue to integrate imaging, pathology, clinical records and genomics data into a composite system. However, this vision contrasts sharply with modern medicine’s trajectory towards more granular sub-specialization. This tension, coupled with data scarcity, domain heterogeneity and limited interpretability, creates a gap between benchmark success and real-world clinical value. We argue that the immediate role of FMs lies in augmenting, not replacing, clinical expertise. To separate hype from reality, we introduce real-world evaluation and assessment of FMs (REAL-FM), a multi-dimensional framework assessing data, technical readiness, clinical value, workflow integration and responsible artificial intelligence. Using REAL-FM, we find that although FMs excel in pattern recognition they fall short on causal reasoning, domain robustness and safety. Clinical translation is hindered by scarce representative data for model training, unverified generalization beyond over-simplified benchmark settings and a lack of prospective outcome-based validation. This Perspective provides clinicians with a practical way to interpret FM claims, identify where these systems may safely support imaging workflows and recognize why human oversight remains indispensable. For developers, it defines the validation, workflow, safety and governance requirements that must be met before FMs can become clinically reliable tools. We envision that the path forward lies not in a monolithic medical oracle, but in coordinated subspecialist AI systems that are transparent, safe and clinically grounded. This Perspective assesses the role of foundation models as clinical tools and introduces real-world evaluation and assessment of foundation models (REAL-FM), a framework for assessing data, technical readiness and clinical value towards responsible artificial intelligence.