Microsoft’s medical artificial intelligence model, MAI-DxO, has demonstrated promising performance in diagnosing complex diseases, according to research highlighted by the company. The system is designed to support doctors in evaluating difficult cases rather than replace human medical professionals.
According to Microsoft, the model was tested on complex medical cases in which reaching a diagnosis required combining patients’ symptoms, medical histories and test results. The company reported that the system outperformed human doctors in certain diagnostic evaluations and, in some cases, reached conclusions with fewer tests and lower costs.
The research also included challenging cases associated with emergency departments in the United States. In one specific diagnostic evaluation, the AI system reportedly achieved approximately 1.7 times the accuracy of doctors.
In another comparison, MAI-DxO recorded a diagnostic accuracy rate of 83.2%, compared with 36.1% for human doctors. In a separate evaluation involving emergency cases, the model achieved 80.3% accuracy, while doctors recorded 46.8%.
Microsoft also examined the number and cost of medical tests required to reach a diagnosis. The company said that, in some cases, the AI system helped reach accurate conclusions at approximately half or one-third of the cost associated with the doctors’ approach.
However, these findings relate to specific cases and experimental conditions. The reported results should not automatically be interpreted as evidence that the system will outperform doctors across all diseases, patients or real-world clinical settings.
Medical AI systems could potentially help physicians organise patient information, compare possible diagnoses and identify situations in which additional testing may be unnecessary. Nevertheless, further research and validation are needed to establish their safety, reliability and effectiveness in everyday clinical practice.
The findings highlight the growing role of artificial intelligence in healthcare. For now, systems such as MAI-DxO are best viewed as potential diagnostic support tools that could assist medical professionals, rather than technologies that can independently replace doctors.