MAI-DxO: Sequential Diagnosis with Language Models: AI Diagnostic System Achieves 80% Accuracy, Reduces Costs by 20%
In June 2025, Microsoft team proposed MAI-DxO diagnostic orchestrator, building a sequential diagnosis benchmark based on 304 NEJM challenging cases. MAI-DxO combined with o3 model achieves 80% diagnostic accuracy (general physicians 20%), with a 20% cost reduction. When configured for maximum accuracy, it reaches 85.5%, and generalizes to models such as Gemini, Claude, DeepSeek.
MAI-DxO simulates the iterative diagnostic process of physicians, upgrading static Q&A to dynamic diagnostic decision-making, significantly enhancing the practicality of LLMs in real clinical scenarios. Its 80% accuracy far exceeds the average 20% of physicians, with lower costs, demonstrating the great potential of AI in assisted diagnosis. This work also provides a standardized sequential diagnosis benchmark, expected to shift the medical AI evaluation paradigm from static to dynamic.
MAI-DxO adopts a model-agnostic orchestrator simulating multi-physician consultation: first, the LLM generates a differential diagnosis list, then selects the next test through information gain and cost-effectiveness analysis. A gating model releases findings only upon explicit query, avoiding information leakage. On 304 NEJM-CPC cases, o3+MAI-DxO achieves 80% accuracy, while general physicians only 20%. Ablation experiments show that iterative strategy and cost-aware selection are key. The system can adapt to various LLM backends with strong robustness.
This technology can be directly embedded into electronic medical record systems to assist primary care physicians in diagnosing challenging cases, reducing misdiagnosis rates and referral costs. If Microsoft productizes it (e.g., Azure Health Bot), it will capture the medical AI decision market. Meanwhile, the sequential diagnosis benchmark may become a new standard for regulatory agencies like FDA to evaluate AI diagnostic tools.
Recommend medical IT suppliers to collaborate with Microsoft to integrate MAI-DxO into clinical decision support systems, prioritizing deployment in emergency departments and general outpatient clinics. Investment should focus on Microsoft's medical AI division developments and third-party diagnostic services based on this technology.
Focus on prospective clinical trial results of MAI-DxO in real hospital settings, as well as its ability to handle rare diseases and multi-morbidity scenarios. It is necessary to verify whether cost reduction is sustainable and the adoption rate of AI suggestions by physicians. If open-sourced, it will accelerate iteration for medical AI startups.