According to PYMNTS.com, a massive study published in Nature Cancer used real-world data from over 15,000 patients across 38 tumor types to test multimodal AI’s predictive power. The model successfully identified key factors influencing survival and treatment response by analyzing medical images, clinical notes and tumor biology together. In colorectal cancer research, the AI system improved prediction accuracy about which patients need chemotherapy after surgery. At this year’s ASCO conference, researchers presented another model for high-risk prostate cancer that integrates pathology images, genetic markers and patient histories. Companies like BostonGene and Flatiron Health are already developing these tools for real-world use, with BostonGene’s platform creating personalized “molecular portraits” of tumors.
Why this actually matters
Here’s the thing – we’ve been hearing about AI in medicine for years, but most of it has been single-purpose stuff. An algorithm that reads X-rays, another that analyzes genetic data. But cancer doesn’t work that way. It’s this incredibly complex interplay between your genetics, your body’s composition, how your immune system responds, and the tumor’s own behavior.
What makes multimodal AI different is it’s finally looking at the whole picture. It’s connecting dots that human doctors simply can’t see because the relationships are too subtle or the data too scattered across different hospital systems. We’re talking about being able to tell a stage II colorectal cancer patient with confidence, “You don’t need chemotherapy” – sparing them months of brutal side effects. Or identifying which prostate cancer patients will actually benefit from expensive second-generation hormone therapies.
The business race is already on
BostonGene isn’t just doing research – they’re building an actual platform that combines genomic, transcriptomic, proteomic and digital pathology data. That’s the business model: creating these comprehensive “molecular portraits” that doctors can use to tailor therapies. Meanwhile, Flatiron Health is tackling the data problem head-on by using large language models to extract meaningful information from the messy, unstructured text in patient records.
Think about the revenue potential here. Better treatment matching means fewer wasted therapies, which saves insurance companies and healthcare systems money. More precise clinical trials mean drug development gets cheaper and faster. And let’s be real – any technology that can demonstrably improve cancer outcomes is going to command premium pricing.
But here’s the catch
The technology might be ready, but our healthcare system isn’t. Medical data lives in these isolated silos – different hospitals use different electronic records systems that don’t talk to each other. Privacy regulations, while important, create another layer of complexity. Even anonymized data sharing faces hurdles.
And the FDA is watching closely. Their January 2025 draft guidance shows they’re taking AI in healthcare seriously, requiring evaluation across the entire lifecycle of these tools. That’s good for safety, but it means companies can’t just deploy these systems and hope for the best. They need robust validation and ongoing monitoring.
So where does this leave us? We’re at that inflection point where the research is compelling enough that real companies are building real products. The potential to make cancer treatment both more effective and more cost-efficient is enormous. But we’re still years away from your local hospital having these tools as standard equipment. The race isn’t just about building better AI – it’s about solving the messy reality of healthcare data and regulation.
