Microsoft tests AI system to support medical diagnosis
Microsoft has presented results from an AI system tested on simulated clinical cases. The company reports better performance than physicians under the evaluation conditions, but the results do not by themselves demonstrate benefits for patients.
Summary
Microsoft has presented an evaluation of MAI-DxO, an AI system designed to work through medical cases step by step. The company reports that the system achieved greater diagnostic accuracy than physicians under the specific test conditions.
For complex cases published in the *New England Journal of Medicine*, Microsoft reports 83% diagnostic accuracy for the system and 36% for physicians with access to reference resources. The comparison uses evaluation cases and does not by itself demonstrate improved patient care.
In practice
The system coordinates several AI perspectives to propose hypotheses, choose questions or tests, and revise its diagnosis as information becomes available. The evaluation considers not only the final answer but also the path taken and the estimated cost of tests.
Microsoft used two sets of cases: complex cases from the *New England Journal of Medicine* and more typical emergency-department presentations. For the second set, the company reports 80% accuracy for the system and 47% for physicians.
What we still don't know
Microsoft says the work was published in *Science*. The paper’s page was inaccessible during this research, so the figures above are attributed to the company’s presentation.
The results need to be considered alongside the exact test conditions, including the resources available to participants. Testing on historical cases is not a substitute for prospective clinical evaluation with patients.
Why it matters
- The study explores AI support for a diagnostic process, beyond answering a single question.
- The comparison depends on the cases and conditions provided to physicians and the system.
- Safety, clinical integration and patient benefits require further evaluation.