Microsoft’s MAI-DxO beats physicians in controlled diagnostic benchmarks
A Science study reports that Microsoft’s MAI-DxO system achieved higher diagnostic accuracy than participating generalist physicians across two controlled datasets; the findings are not evidence of real-world clinical deployment.
AT A GLANCE
Key points
- MAI-DxO reached 83% accuracy on 304 complex NEJM cases in the study.
- On MIMIC-IV-ED cases, the system achieved 80% accuracy versus 47% for participating physicians.
- The findings are benchmark results, not evidence of approved clinical use.
- Microsoft plans prospective evaluation with Beth Israel Deaconess Medical Center.
What is changing?
Microsoft AI evaluated its model-agnostic MAI-DxO orchestrator with SDBench, a benchmark for sequential diagnosis. The Science paper reports 83% accuracy on 304 difficult NEJM cases, compared with 22% for physicians without external references and 36% when references were allowed.
On the more typical emergency-department presentations in MIMIC-IV-ED, MAI-DxO reached 80% accuracy versus 47% for physicians. Instead of producing a single answer, the system acts like a virtual panel that proposes questions and tests, updates probabilities as evidence arrives, and considers test cost.
The results come from two controlled datasets; they do not constitute clinical approval or prove superiority in real hospitals. Microsoft says a planned prospective study with Beth Israel Deaconess Medical Center will test the system in real care settings using independent physician assessments.
Explore the original sources
Source date: 22 October 2026.
Published by AI Radar: 23 Oct 2026, 10:30 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…