Science & Tech · AI · 9 hrs ago
Microsoft AI tool diagnoses complex cases more accurately than generalist doctors
Microsoft’s DxO system diagnosed 83% of complex cases correctly in a simulation, a study published in Science found.
The cases came from the New England Journal of Medicine, and DxO uses several AI agents to ask questions, request tests and discuss possible diagnoses.
Generalist doctors using reference tools scored 36% in the comparison; without those tools, their average was 22%.
The results do not show how DxO compares with specialists or teams of doctors, who often handle cases this complex.
The findings suggest AI could help with difficult diagnoses, but performance varied by the underlying AI model and case type, with less benefit on emergency department cases.
Microsoft says it is working with Mayo Clinic on a healthcare AI model, with a release planned for next year.
Microsoft’s MAI-DxO diagnostic system correctly diagnosed 83.2% of complex medical cases in a simulation, compared with 36.1% for generalist physicians using search and reference tools.
The study was published in Science on Oct. 22.
The cases came from the New England Journal of Medicine, and the comparison did not include specialists or physician teams.
MAI-DxO coordinates several AI agents that question patients, order tests and debate diagnoses.
Microsoft is working with Mayo Clinic on a healthcare foundation model, with release planned for next year.
- Who
- Microsoft’s MAI-DxO system and generalist physicians took part in the comparison.
- What
- In a simulation, the AI diagnosed complex cases more accurately than generalist physicians using search and reference tools.
- When
- The study was published Oct. 22, 2026.
- Where
- The cases were from the New England Journal of Medicine; the study’s setting is not stated.
- Why
- Not stated.
This story does not have two clearly opposing sides.
The study comparing MAI-DxO with generalist physicians was published in Science.
Microsoft plans to release a healthcare foundation model it is developing with Mayo Clinic.
Mustafa Suleyman said Microsoft could be on the verge of a breakthrough in medical-record coding.
- AI system
- MAI-DxO, also called Diagnostic Orchestrator or DxO
- AI accuracy
- 83.2% on complex cases
- Physician accuracy
- 36.1% using search and reference tools; 22.3% without them
- Case source
- New England Journal of Medicine
- Study publication
- Oct. 22, 2026
3 sources
What people are saying
medical superintelligence
I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records.
The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise.