Diagnosing Like a Panel of Experts
According to the report, the model uses an “orchestrator” that organises how AI tackles a problem, employing a simulated panel of clinical experts collaborating to solve diagnostic cases. Existing medical AI models typically take an all-at-once approach, examining all of a patient’s symptoms, lab results and imaging studies to generate a diagnosis.
The new system, by contrast, mimics how doctors actually think. Throughout the process, it keeps updating its working hypotheses and plans, the researchers said. The main steps are:
- Deciding what to ask about the patient’s history and what is significant in the physical exam.
- Choosing which tests to order based on what it has learned.
- Considering the results, ordering further testing as needed and refining the diagnosis.
Twice as Accurate as General Physicians
Tasked with diagnosing more than 300 highly unusual and confusing cases, the model using the orchestrator was more than twice as accurate as general physicians using textbooks and the internet, both in working up the patient and in reaching the diagnosis. The tool also outperformed doctors on scenarios drawn from a major database of emergency room admissions.
















