Researchers at the University at Buffalo in New York have built an AI-based system that can distinguish human-written radiology reports from those generated by large language models, aiming to prevent medical fraud and protect insurance workflows.
The research team developed a dataset of 14,000 paired chest X-ray reports — one human-authored and one AI-generated — to train and test the detection model, according to a March 16 news release. The tool achieved up to 100% accuracy across categories and maintained high performance when tested against unfamiliar language models.
Unlike general-purpose AI detectors, the Buffalo system was designed specifically for radiology’s domain-specific language and stylistic norms. The model uses style disentanglement techniques to separate clinical content from subtle patterns in phrasing and word choice that reveal nonhuman authorship.
The team said the work has implications beyond medicine, offering a potential safeguard for any field vulnerable to AI-generated forgeries, including insurance, law and journalism. The researchers plan to expand the tool to additional specialties and release the benchmark dataset for public use.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
