A research team from Mass General Brigham has built one of the first fully independent artificial intelligence (AI) systems that can detect signs of cognitive problems by reading regular medical notes.
Once it is set up, this system works on its own and doesn’t need help from doctors or nurses. In real-world tests, it correctly ruled out people without cognitive issues 98% of the time. The results were recently published in the journal npj Digital Medicine.
This AI tool is important because many people with cognitive decline, such as memory loss or confusion, don’t get diagnosed early. Traditional screening tests are time-consuming and hard for patients to access. Early diagnosis is more important than ever because there are now new Alzheimer’s disease treatments that work best if given early.
The new AI system, created by Dr. Hossein Estiri and his team, is made up of five smaller AI tools called “agents.” Each one has a special job.
They work together and even review each other’s work—just like a team of doctors discussing a patient. This kind of teamwork helps make sure the system thinks carefully and checks for mistakes before making a decision.
The researchers tested their system on over 3,300 clinical notes from 200 anonymous patients. These notes were written by doctors during regular visits, not special memory tests. The AI was able to find patterns and clues in these notes that might suggest a person is experiencing cognitive decline.
As one researcher explained, regular clinical notes often include “whispers” of brain-related issues that busy doctors don’t always have time to pick up. This system listens carefully and can spot early signs at scale.
One of the best parts of this system is that it can run completely within a hospital’s computer system. It does not send patient information to outside servers or the internet, which helps protect patient privacy.
In cases where the AI and human reviewers disagreed, the researchers brought in an independent expert to review the situation. In over half of these cases, the expert agreed with the AI. That means the AI wasn’t just guessing—it was often making well-reasoned decisions based on the evidence in the notes.
Still, the AI is not perfect. It had more trouble when clinical notes were missing context, or when cognitive concerns were only briefly mentioned. It also struggled with certain types of medical language it wasn’t trained to understand. The system performed very well when the notes were clear and complete.
In balanced test conditions, the AI system correctly identified 91% of people with cognitive problems. But in real-life situations—where only about one-third of people had cognitive issues—its sensitivity dropped to 62%, though its ability to correctly rule out people without problems stayed high at 98%.
The team says they are being open about these weaknesses so other researchers can help improve the system. They’ve even released a tool called Pythia, which lets hospitals and researchers use the AI system and adjust it for their own needs.
This research offers new hope for catching cognitive decline earlier by using tools already built into the healthcare system. It could help more people get diagnosed early and receive treatment before it’s too late.
If you care about dementia, please read studies that eating apples and tea could keep dementia at bay, and Olive oil: a daily dose for better brain health.
For more health information, please see recent studies what you eat together may affect your dementia risk, and time-restricted eating: a simple way to fight aging and cancer.
Copyright © 2025 Knowridge Science Report. All rights reserved.
