Microsoft Develops AI Tool for Sequential Medical Diagnosis
Microsoft has developed an artificial intelligence system that approaches medical diagnosis more like a doctor. Called MAI-DxO, the system gathers patient information, requests tests, and revises its conclusions as new evidence emerges. A study published in Science on October 22, 2026, reports promising results across complex medical cases and emergency department records.
Unlike conventional AI tools that analyse a complete medical case at once, MAI-DxO follows a sequential process. First, it reviews the initial symptoms. Then, it decides which questions to ask and which tests to request. Finally, it updates its diagnostic hypotheses as the evidence develops.
How Microsoft’s AI Diagnostic Orchestrator Works
MAI-DxO coordinates several AI-driven roles that simulate a panel of medical professionals. These roles propose possible diagnoses, challenge assumptions, and identify the next useful investigation. Meanwhile, an orchestrator evaluates the options and weighs diagnostic value against estimated testing costs.
Microsoft researchers tested the system using two datasets. The first contained 304 complex cases from The New England Journal of Medicine. The second included 325 emergency department encounters drawn from the MIMIC-IV-ED database.
The researchers also created the Sequential Diagnosis Benchmark (SDBench) to evaluate how AI systems gather evidence and make decisions. In this benchmark, the system receives information only when it requests specific details or tests. As a result, researchers can assess the diagnostic process, not just whether the final answer is correct.
Study Reports Higher Diagnostic Accuracy
On complex medical cases, MAI-DxO achieved 83.2% diagnostic accuracy. By comparison, generalist physicians reached 36.1% when they could consult textbooks and online resources. Without those resources, physicians scored 22.3%.
The system also performed well on more typical emergency department cases. It achieved 80.3% accuracy, compared with 46.8% for physicians in the study. However, these results came from controlled evaluations rather than live clinical practice.
Nevertheless, the findings require careful interpretation. The study compared the system with individual generalist physicians, rather than specialist teams. It also relied on text-based cases and estimated testing prices. Moreover, the researchers did not evaluate treatment decisions or patient outcomes.
Microsoft Plans Real-World Clinical Testing
Microsoft plans to evaluate MAI-DxO with Beth Israel Deaconess Medical Center in Boston. The collaboration will examine how the system performs in clinical settings and whether it can support patient triage. Researchers will also assess how its recommendations compare with physicians’ independent evaluations.
The company sees the project as part of a broader effort to develop advanced medical AI. Ultimately, it aims to support clinicians with diagnostic reasoning, evidence gathering, and more efficient investigations.
However, the current findings do not establish that AI can replace doctors. Instead, they show that sequential diagnostic systems may offer useful support under controlled conditions. Further clinical testing will determine whether these gains translate into safer, more effective patient care.
