Researchers at Mass General Brigham, the oldest and largest teaching affiliate of Harvard Medical School, have launched an autonomous “agentic” AI workflow that identifies early signs of cognitive impairment with 98% specificity. Published in npj Digital Medicine, the system functions like a “digital clinical team,” autonomously sifting through routine medical notes to surface signs of dementia that busy clinicians often overlook.
The breakthrough lies in its multi-agent architecture. Rather than using a single prompt, the system employs five specialized AI agents simulating a professional case conference. In real-world testing on over 3,000 clinical notes, according to the report, the AI matched expert-level accuracy and even correctly identified clinical concerns that initial human reviews had missed.
By running on open-weight models within secure hospital servers, the tool will ensure patient privacy while closing the critical diagnosis gap. The researchers have open-sourced the tool, named “Pythia”, to help global healthcare providers identify at-risk patients early enough to benefit from time-sensitive Alzheimer’s treatments.
