Foundation Models in Biomedical Imaging: Turning Hype into Reality

Amgad Muneer1, Kai Zhang1, Ibraheem Hamdi2, Rizwan Qureshi 3, Muhammad Waqas1, Shereen Fouad4, Hazrat Ali5, Syed Muhammad Anwar 6,7, Jia Wu 1,8

1 Department of Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA 2 Center for Secure Artificial Intelligence For hEalthcare (SAFE), McWilliams School of Biomedical Informatics, UTHealth Houston, TX, USA 3 Female Medicine in Machine Learning, Massachusetts Institute of Technology, MA, USA 4 School of Computer Science and Digital Technologies, Aston Centre for Artificial Intelligence Research and Application, Aston University, UK 5 Division of Computing Science and Mathematics, University of Stirling, Stirling, UK 6 School of Medicine and Health Sciences, George Washington University, Washington, DC, 20052, USA 7 Sheikh Zayed Institute, Childrens National Hospital, Washington, DC, 20010, USA 8 Department of Thoracic/Head and Neck Medical Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA Abstract Foundation models (FMs) are driving a prominent shift in artificial intelligence across different domains, including biomedical imaging. These models are designed to move beyond narrow pattern recognition towards emulating sophisticated clinical reasoning, understanding complex spatial relationships, and integrating multimodal data with unprecedented flexibility. However, a critical gap exists between this potential and the current reality, where the clinical evaluation and deployment of FMs are hampered by significant challenges. Herein, we critically assess the current state-of-the-art, analyzing hype by examining the core capabilities and limitations of FMs in the biomedical domain. We also provide a taxonomy of reasoning, ranging from emulated sequential logic and spatial understanding to the integration of explicit symbolic knowledge, to evaluate whether these models exhibit genuine cognition or merely mimic surface- level patterns. We argue that a critical frontier lies beyond statistical correlation, in the pursuit of causal inference, which is essential for building robust models that understand cause and effect. Furthermore, we discuss the paramount issues in deployment stemming from trustworthiness, bias, and safety, dissecting the challenges of algorithmic bias, data bias and privacy, and model hallucinations. We also draw attention to the need for more inclusive, rigorous, and clinically relevant validation frameworks to ensure their safe and ethical application. We conclude that while the vision of autonomous AI-doctors remains distant, the immediate reality is the emergence of powerful technology and assistive tools that would benefit clinical practice. The future of FMs in medical imaging hinges not on scale alone, but on developing hybrid, causally aware, and verifiably safe systems that augment, rather than replace, human expertise, and in our study, we found that the field is gradually moving towards this direction. Keywords: Foundation models, biomedical imaging, medical AI, clinical reasoning, causality, trustworthiness, and safety.

1. Introduction
   The development and deployment of artificial intelligence (AI) is undergoing a paradigm shift, moving from an era of models trained on relatively small datasets for specific tasks to a dominant foundation model (FM) capable of multiple tasks. In this pursuit, there are two fundamental innovations in this new paradigm:

1. model training on big data with an aim to perform well in a wide variety of downstream tasks; 2) model training is usually unsupervised and does not rely on the available ground truth labels. Such large-scale models, pretrained on vast and diverse datasets, have been adapted to a multitude of downstream applications with remarkable flexibility.1,2 This paradigm shift also holds profound implications for biomedical engineering in general and particularly medical imaging, a field defined by its complex, multimodal data, where getting high-quality ground truth labels at scale is a big challenge3. Specialized AI methods have already demonstrated success in discrete tasks like identifying diabetic retinopathy 4,
   segmenting tumors5 or detecting breast cancer 6 using medical images. However, FMs promise a more holistic, generalist intelligence capable of integrating diverse data sources (both structured and unstructured) such as imaging, electronic health records, genomics, and clinical notes to emulate the comprehensive reasoning of a human expert.3
   This potential of current AI technology has generated immense excitement, heralding a future of augmented diagnostics, personalized treatment planning, precision medicine, and streamlined clinical workflows.7–9 However, it has also fueled considerable hype, creating a critical need to distinguish between the future potential of these models and their