Microsoft's MAI-DxO AI System Achieves 85.5% Diagnostic Accuracy

Microsoft has made significant strides in the field of healthcare AI with the introduction of MAI-DxO, a system designed to simulate multiple virtual doctors working collaboratively to solve complex medical cases. The system, announced by Microsoft CEO Satya Nadella, has shown impressive results in diagnostic accuracy and cost efficiency.

In tests conducted against 304 complex cases from the New England Journal of Medicine, MAI-DxO achieved an 85.5% diagnostic accuracy rate. This is a substantial improvement over a group of 21 experienced physicians, who correctly diagnosed only 20% of the same cases. The AI system not only outperformed human doctors in accuracy but also demonstrated cost savings, spending approximately 20% less on diagnostics in one configuration.

MAI-DxO operates through the Sequential Diagnosis Benchmark (SDBench), which mimics the real-world diagnostic process. The system starts with limited patient information, asks follow-up questions, orders tests, and adjusts theories as new data becomes available. Each test incurs a cost, forcing the AI to balance thoroughness with healthcare spending. The virtual physician panel includes roles such as Dr. Hypothesis, who maintains a list of the most likely diagnoses, and Dr. Test-Chooser, who selects diagnostic tests for maximum information gain.

The system was tested on cases published in the New England Journal of Medicine between 2024 and 2025, ensuring that the AI had not memorized the answers. The 21 physicians recruited for comparison had between 5 and 20 years of experience and worked without access to colleagues, textbooks, or AI assistance.

Microsoft's broader push into consumer health AI includes over 50 million health-related sessions daily across its Bing and Copilot products. The company envisions search engines and AI assistants becoming the new front door for healthcare. MAI-DxO is designed to work with AI models from different companies, boosting performance across various models by an average of 11%.

Despite its promising results, Microsoft emphasizes that MAI-DxO is still in the research phase. The technology needs further testing on routine cases and rigorous safety testing, clinical validation, and regulatory reviews before it can be deployed in clinical settings. The company plans to submit the research for peer review and collaborate with healthcare organizations to validate the approach.

While MAI-DxO represents a significant step toward AI-assisted healthcare, it is still too early to see mainstream implementation. Microsoft's vision is for AI to augment rather than replace human doctors, providing a valuable tool to improve diagnostic accuracy and reduce costs. The 21 physicians who scored 20% on the NEJM cases are likely hoping that AI will indeed augment their capabilities in the future.