What is it about?
Artificial intelligence (AI) is transforming how doctors diagnose and treat cancer. This paper reviews the breakthrough performance and clinical potential of 'MUSK,' an advanced AI foundation model published in the journal Nature (January 2025). MUSK is uniquely trained on over 50 million cancer tissue images and 1 billion clinical text tokens. Unlike previous AI models that require expensive genetic testing or high-tech equipment, MUSK works with routinely available patient data—standard tissue slides and everyday clinical notes. It has successfully outperformed traditional systems in predicting cancer recurrence and immunotherapy responses. Because it relies on basic, standard medical records, this AI can be easily deployed in low-resource areas and developing countries, paving the way for more equitable global cancer care. This paper highlights how MUSK bridges the gap between advanced technology and real-world oncology clinics.
Featured Image
Photo by Ritu Chauhan on Unsplash
Why is it important?
This study is highly timely as it demonstrates the practical clinical translation of medical AI, distinguishing itself from contemporaneous models through two key breakthroughs: First, it operates on routinely available data rather than high-cost diagnostics. Traditional oncology AI models rely heavily on expensive next-generation sequencing (NGS) or highly standardized imaging setup. In contrast, MUSK achieves superior predictive accuracy using only standard, routinely available tissue slides (H&E stained) and unstructured clinical notes. This unique flexibility allows for immediate deployment in low- and middle-income countries, addressing global healthcare equity. Second, it represents a fully vetted, prospective-ready framework published in Nature (January 2025). While many AI models remain at the preprint or prototype stage, MUSK’s methodological rigor and multi-institutional validation have been formally peer-reviewed, affirming its readiness for real-world prospective clinical research. By bridging the gap between advanced self-supervised learning and everyday clinical realities, this work provides an accessible and scalable blueprint for the next generation of precision oncology worldwide.
Perspectives
While working on medical AI, the question that constantly occupied my mind was: "How can we ensure that this advanced technology truly benefits every patient, regardless of where they live?" Even the most brilliant AI loses its value if it is only accessible to elite hospitals with multi-million dollar budgets for genetic sequencing. It was deeply rewarding to explore a framework that challenges this barrier by operating on everyday clinical records and routine tissue slides found in any local hospital worldwide. I hope this article provides food for thought not just for AI researchers, but also for clinicians and policymakers on the front lines. Ultimately, I want this work to show that creating an equitable, accessible future for precision oncology is not a distant dream, but a practical reality we can build today.
Wonbeak Yoo
Korea Research Institute of Bioscience and Biotechnology
Read the Original
This page is a summary of: Translating multimodal foundation models into oncology: Toward a future where AI directs diagnosis and therapy, Genes & Diseases, July 2026, Tsinghua University Press,
DOI: 10.1016/j.gendis.2025.101958.
You can read the full text:
Contributors
The following have contributed to this page
