The Microsoft AI health team posted on October 22, 2026 about a diagnostic system called MAI-DxO. The post says the work was published that day in Science. The graphic reads “1.7 times more accurate than physicians.” The prose puts that multiple in a narrower comparison: on common presentations to a US emergency department, evaluated with cases from MIMIC-IV-ED, MAI-DxO is 1.7 times as accurate as physicians. That is the post's claim. The post does not say how many physicians were in the comparison.
[1]
A second comparison uses complex cases published in The New England Journal of Medicine. The post says that on those cases MAI-DxO is more than twice as accurate as physicians using textbooks and the internet, and nearly four times as accurate as physicians diagnosing without those resources. Cost is also the post's claim: it does this at half or even one third of the cost, and reaches the diagnosis with fewer tests and procedures. The post does not list which tests were skipped, and it does not show the cost ledger. Those sentences are not rewritten here as a diagnostic procedure.
[1]The same week, the post points to a vision piece in BMJ Digital Health and AI. That piece's measurements are not in this post. The post opens with a question about accuracy, cost, and the number of tests. The question is not a result. The results on the page are the comparisons above and the Science publication. The post does not say whether the system is already in use in hospitals, and it does not say which year's MIMIC-IV-ED cases were used.
[1]