Microsoft says its MAI-DxO AI diagnoses patients better than doctors

Microsoft's MAI-DxO model reached 83.2% diagnostic accuracy on 304 NEJM clinicopathological cases versus 36.1% for generalist physicians in a peer-reviewed study in Science. On 325 MIMIC-IV-ED emergency cases it scored 80.3% against 46.8%, at half to one-third of the doctors' testing costs.

• 83.2% accuracy on 304 NEJM cases vs 36.1% for physicians
• 80.3% vs 46.8% on 325 MIMIC-IV-ED emergency cases
• Testing costs half to one-third of human doctors'
• Model gathers information sequentially, ordering tests like a doctor