Microsoft's diagnostic AI study is published in Science, and clinical testing is planned in Boston

Microsoft's study of its AI Diagnostic Orchestrator, MAI-DxO, has been published in the peer-reviewed journal Science. The paper, titled "Sequential Diagnosis with Language Models," was first posted on arXiv last year as a preprint that had not been peer reviewed. Microsoft says the system "substantially outperformed practicing generalist physicians" in diagnostic accuracy while ordering fewer tests and procedures. On complex cases published in The New England Journal of Medicine, MAI-DxO reached 83.2 percent accuracy. Generalist physicians with access to reference materials reached 36.1 percent. Without those resources, physician accuracy fell to roughly 22 percent. In a separate test on emergency department admissions, the system also beat generalist physicians, though by a narrower margin. Microsoft AI chief Mustafa Suleyman told The Verge that publication in a peer-reviewed journal is a major step toward validating the results, but that any real-world deployment would still require clinical testing. Dominic King, Microsoft's AI health lead, said testing will begin at Beth Israel Deaconess Medical Center in Boston within the next couple of months. The results come from benchmark cases, not live patient care. The planned clinical evaluation will be the first test of whether the system's performance holds up in a hospital setting.