DeepRare surpasses physicians in a study on diagnosing rare diseases.

      DeepRare, an AI system with 40 specialized tools, surpassed medical experts in detecting rare conditions in a direct study published in Nature.

      For millions with rare diseases, the journey to a proper diagnosis is complex and convoluted. Patients often navigate between general practitioners and specialists for years, sometimes decades, trying to identify symptoms that are not typically documented.

      Eighty percent of rare diseases have a genetic basis, yet many cases remain undiagnosed until significant biological damage has occurred. The challenge lies not in the availability of data, but in identifying critical information within an overwhelming amount of medical data.

      A recent study released in Nature indicates that artificial intelligence could speed up this search. Researchers from Shanghai Jiao Tong University’s School of Artificial Intelligence and Xinhua Hospital developed DeepRare, an AI system aimed at emulating the diagnostic reasoning used by human doctors when faced with uncertainty.

      In a direct comparison involving five skilled physicians, each with over ten years of experience, the AI achieved consistently greater accuracy.

      The results are impressive. DeepRare accurately identified the disease on its first attempt 64.4 percent of the time, while the doctors achieved a 54.6 percent accuracy rate. When provided with three suggestions instead of one, the AI managed a diagnostic success rate of 79 percent compared to 66 percent for the human doctors.

      Importantly, physicians supported the AI’s reasoning 95.4 percent of the time, indicating that not only does the system arrive at correct conclusions, but it also does so in a manner that seasoned clinicians find convincing and clinically valid.

      What sets DeepRare apart from previous diagnostic AIs is its framework. Rather than using a black-box classification model, it merges 40 specialized digital tools and adheres to a clearly reasoned workflow.

      It generates diagnostic hypotheses, validates them against patient data, consults global medical literature, assesses genetic variants, and continuously refines its conclusions before ranking the potential diagnoses.

      This approach reflects the cognitive processes employed by human diagnosticians, but with the advantage of access to comprehensive medical knowledge and computational abilities beyond human capacity.

      DeepRare is no longer confined to the lab. Since July 2025, it has been implemented on an online diagnostic platform, with over 600 medical institutions worldwide registered to utilize it.

      The research team intends to further validate the system using 20,000 actual cases and to establish a global alliance focused on rare disease diagnostics. Importantly, the authors stress that this system is designed to assist, not replace, clinicians, recognizing both the limitations of AI technology and the essential human element within the medical field.

      The consequences for patients are significant. Approximately 300 million individuals globally suffer from rare diseases, with the average diagnostic journey extending to five years or more.

      Every year of delay in obtaining a diagnosis contributes to uncertainty, incorrect treatments, and worsening health. An AI system capable of reducing this timeline by weeks or months and uncovering potential diagnoses that might be missed could transform the initial experience of living with a rare disease.