Microsoft researchers published a study of the Diagnostic Orchestrator (DxO), also called Microsoft AI Diagnostic Orchestrator (MAI-DxO), in Science on 22 October 2026. The system coordinates diagnostic steps for complex medical cases, including asking questions, ordering tests and revising possible diagnoses.
Clinical Decision Support and Diagnostic Process
Clinical decision-support systems use patient information to assist healthcare professionals with decisions such as diagnosis and treatment. MAI-DxO simulates a panel of physicians by carrying out a sequence of diagnostic actions rather than producing a diagnosis in a single step. Its orchestration layer can work with large language models, including OpenAI’s GPT-5 and GPT-5.5. In reported testing, the orchestration layer raised GPT-5.5’s diagnostic accuracy from 72% to nearly 87%.
Diagnostic Accuracy and Case Sets
The study tested DxO on 304 complex clinicopathological conference cases from the New England Journal of Medicine. It achieved 83.2% diagnostic accuracy, compared with 36.1% for generalist physicians using search and reference tools. On a separate set of 325 emergency department admissions, DxO achieved 80.3% accuracy, while physicians scored 46.8%. These results relate to the tested case sets and do not constitute a measure of performance across all clinical settings.
Sequential Diagnosis Benchmark and Cost Data
Microsoft is open-sourcing SDBench, the Sequential Diagnosis Benchmark used in the study, as a common benchmark for comparing diagnostic systems. In simulated cases, an earlier cost analysis found that the tool reduced estimated diagnostic costs by 20% compared with physicians and by 70% compared with off-the-shelf OpenAI o3 models. In the 2026 figures, DxO averaged 5,431 per complex case without budget limits and 2,241 with budget constraints; the constrained setting achieved 80.9% accuracy. Physicians averaged $4,725 per case.
Important Facts for Exams
- The Diagnostic Orchestrator is an artificial-intelligence system designed to coordinate sequential diagnostic actions.
- The New England Journal of Medicine publishes clinicopathological cases used in medical education and diagnostic discussions.
- SDBench stands for Sequential Diagnosis Benchmark and provides a benchmark for evaluating sequential diagnosis systems.
- Microsoft plans to test the tool for emergency department waiting-room triage at Beth Israel Deaconess Medical Center in Boston.
Microsoft’s planned hospital testing is intended to assess the tool for triaging patients in emergency department waiting rooms. The reported diagnostic accuracy figures come from case-based evaluations and simulated testing.
