Building Effective AI Foundation Models at Standard Model Biomedicine With Kevin Brown and Waqas Haque, MD, MPH

What Are Foundation Models?

Kevin Brown began by explaining the concept of a foundation model in medicine. “A foundation model is a massively pre-trained neural network. You can think of it as a preassembled brain that’s learned patterns across millions of medical records, radiology images, or pathology slides,” he said. Rather than building a new AI tool from scratch for each problem, these models can be fine-tuned for specific tasks, saving time and resources. As Brown put it, “It really democratizes access to state-of-the-art performance.”

Breaking Down Data Silos

One of the biggest hurdles in biomedical AI is the fragmentation of data. Hospitals, labs, and pharmaceutical companies often store information in different formats and platforms, making it difficult to create comprehensive AI models. “Our approach uses federated and privacy-preserving training so that a single foundation model can learn across all of these different silos,” Brown explained. He emphasized that expertise is also siloed, even within oncology, and integrating diverse data types—like radiology, genomics, and clinical notes—is essential for a complete understanding of each patient. “A patient is more than just the sum of the measurements you make. It’s how they all fit together,” he noted.

The True Cost of AI

While AI tools like ChatGPT seem effortless to use, the underlying costs are significant. “Training a large language model from scratch can cost hundreds of millions of dollars, but running these models is much cheaper,” Brown said. This means that once a robust foundation model is built, it can be deployed widely at minimal cost, making advanced AI accessible to more clinicians and researchers.

The Importance of Domain Expertise

AI in medicine isn’t just about data and algorithms—it requires deep clinical knowledge. “Biomedical AI really demands a nuanced understanding, because patient treatment and care require it,” Brown emphasized. Foundation models can absorb broad clinical patterns, but the final, context-specific solutions must be built by domain experts. “That last 10% takes 90% of the effort and expertise,” he observed, highlighting the importance of collaboration between clinicians and AI specialists.

Moving from Reactive to Proactive Care

Standard Model Biomedicine is developing “next event prediction” (NEP) models, which forecast clinical milestones such as tumor progression or adverse events. “Next event prediction is kind of like how ChatGPT predicts the next word, but applied to patients,” Brown explained. This approach could help oncologists anticipate complications and intervene earlier, shifting care from reactive to proactive. “Oncologists could get an early warning signal—for example, triaging in radiation therapy for cardiac adverse events.”

Advice for Clinicians

For oncologists interested in AI, Brown recommends partnering with AI teams at academic centers. “They’ll have the ability to bridge the technical side into the actual unmet needs that the clinicians see,” he advised. With rising cancer cases, AI can help reduce workload and improve patient outcomes, for example by streamlining clinical trial matching.

Looking Ahead

In the coming years, Brown expects to see AI tools running in “shadow mode” alongside clinicians, providing risk profiles and tracking outcomes without directly driving interventions. “This would enable efficient trial design that streamlines and directly addresses patient needs,” he predicted.

Key Takeaways

• Foundation models are pre-trained AI systems that can be adapted for a wide range of medical tasks, saving time and resources
• Data silos and fragmented expertise are major barriers, but federated learning and collaboration can help overcome them
• Training AI models is expensive, but deploying them is relatively cheap, making advanced tools more accessible
• Domain expertise is crucial for building effective AI solutions in medicine
• Next event prediction models can help shift oncology care from reactive to proactive
• Collaboration between clinicians and AI experts is essential for successful implementation

About the Speakers

Waqas Haque, MD, MPH, is a Hematology/Oncology Fellow at the University of Chicago. He recently completed his Internal Medicine Residency at New York University (NYU) in a Clinical Investigator Track. Dr. Haque has balanced his patient care work with a variety of research projects. During his fellowship training at the University of Chicago, he plans to further his work in innovative clinical trial design, value-based care delivery to cancer patients, and clinical investigation.

Kevin Brown is the Founder and CEO of Standard Model Biomedicine, which provides state-of-the-art foundation models in medical imaging, language, genomics, and pathology for biomedical applications. Previously, he led the oncology foundation model development at Bristol Myers Squibb. He has authored numerous publications on computational neuroscience, medical imaging foundation models, and the use of large language models for clinical outcome prediction in top-tier machine learning venues.