Tool extracts disease descriptions from messy medical text without needing to be retrained for each new disease database

Researchers built a method that uses large language models to automatically recognize disease phenotypes mentioned in medical text, without needing custom training for each new medical ontology or dataset. This means hospitals and research teams can apply the same tool across different disease vocabularies and types of medical writing without rebuilding it each time.

Medical text mining has been stuck between two bad choices: build custom tools for each disease database (expensive, doesn't generalize) or use generic AI that doesn't understand medical terminology. This approach removes that tradeoff by letting a single tool work across different medical vocabularies without retraining. The real impact is speed — research teams can now tag disease mentions in new datasets or switch between disease ontologies without months of custom engineering work.