Foundation models in healthcare: a comprehensive review from technical advances to clinical translation

As artificial intelligence (AI) has evolved through a series of discrete leaps, the Foundation model (FM) has demonstrated substantial potential for applications in the medical domain. Built on scalability, multimodal processing, and adaptability to diverse downstream tasks, FMs offer a flexible framework that can be tailored to various clinical needs. Nevertheless, the translation of FMs into clinical practice remains challenged by concerns regarding data privacy and security, bias and fairness, interpretability and sustainability. Therefore, a clinically oriented review is needed not only to summarize current advances and limitations but also to emphasize the clinical relevance, practical significance, and translational implications of FMs in medicine.

This review outlines the development history of AI and introduces the FM basic theory, summarizes recent advances in their medical applications, and examines how FMs may support clinicians, enhance workflow efficiency, and improve patient outcomes. In addition to summarizing existing work, this review places particular emphasis on the clinical relevance, practical significance, and translational challenges of FMs across healthcare. Furthermore, privacy, safety, transparency, computational resources, clinical feasibility and sustainability issues are further discussed. Finally, the future direction of FMs in the medical field was projected.

A central concept of this review is that the clinical translation of FMs requires interdisciplinary collaboration among AI developers, clinicians, and policymakers, supported by careful evaluation frameworks and continuous oversight to ensure clinical benefit and minimize risk.