Carnegie Mellon's machine minds may outsmart cancer
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios
A new AI tool developed by Carnegie Mellon researchers could help doctors identify patients at high risk of cancer — advancing early detection and potentially saving lives.
The big picture: More hospitals and insurance providers are investing in AI to ease clinician burnout and streamline care, from digital scribes to diagnostic systems that accelerate results.
- Cancer is the second-leading cause of death in the U.S. and across Southwestern Pennsylvania, and it's often caught too late for effective treatment.
How it works: CATCH-FM, or "CATch Cancer early with Healthcare Foundation Models," can analyze electronic health records to predict future medical events, including cancer. It flags high-risk patients using past diagnoses, medications and other health factors linked to cancer.
Stunning stat: The team from CMU's School of Computer Science found that 50% to 70% of patients CATCH-FM identified as high risk for lung, liver or pancreatic cancer were later diagnosed with those diseases, says Chenyan Xiong, an associate professor in CMU's Language Technologies Institute who led the initiative.
- Among patients with no prior cancer, the tool correctly flagged about half who were later diagnosed. For those with a history of cancer (excluding the targeted cancers), it identified 70% as high risk.
Zoom in: CMU researchers trained the AI — which works similarly to chatbots like ChatGPT — with access to millions of anonymous records spanning decades from Taiwan's national health database, partnering with a Taiwanese hospital to explore its clinical use.
- Researchers developed the model in-house rather than using an existing large language model to protect patients' private data, says Xiong.
What they're saying: Xiong says the tool isn't meant to replace scans or doctors but to expedite the pre-screening process and help build a stronger base for AI in health care.
- "Doctors and clinicians know what a patient needs more than us," he says. "Cancer detection may be the champion application," but doctors may eventually adapt the tech to a wider range of health needs, he says.
Between the lines: While AI promises health care workers greater efficiency and more time for patients, rapid expansion has raised questions about data privacy, bias, errors and whether regulators can keep pace — prompting some states to write their own rules in the absence of federal guardrails.
The latest: Pennsylvania lawmakers are considering legislation that would require hospitals, insurers and clinicians to disclose how they use AI and ensure a human makes the final call.
What's next: Xiong said the team is in the early stages of pursuing a collaboration with UPMC to expand research efforts in Pittsburgh.