Diagnostic AI reached 1.7 times generalist physicians' accuracy on emergency cases in a controlled benchmark

The Science study compared Microsoft's MAI-DxO with participating generalist physicians on text-based emergency-admission cases. It achieved 80.3% accuracy versus physicians' 46.8%, but at an estimated $4,979 in testing per case versus $3,400. With budget constraints, its accuracy fell to 73.2% and estimated testing cost to $2,634. The researchers say the controlled benchmark does not establish superiority over specialists or real-world clinical practice.