China unveils AI system DeepRare for rare disease diagnosis

Addressing longstanding diagnostic challenges

Diagnosing rare diseases has historically been difficult, especially in regions with limited access to genetic testing. Traditional medical AI systems have also faced trust concerns due to opaque reasoning processes that make their conclusions difficult to verify.

Record-breaking accuracy

DeepRare was developed by researchers from Xinhua Hospital, affiliated with the Shanghai Jiao Tong University (SJTU) School of Medicine and the SJTU School of Artificial Intelligence. Since launching its online diagnostic platform last July, the system has attracted over 1,000 professional users from more than 600 medical and research institutions worldwide.

Testing showed that when relying solely on patients’ clinical phenotypic information without genetic data, DeepRare achieved a first-attempt diagnostic accuracy of 57.18 percent — nearly 24 percentage points higher than the previous leading global model. When genetic data were included, accuracy surpassed 70 percent.

Evidence-based and transparent reasoning

DeepRare integrates real-time access to extensive medical literature and real-world clinical case data. Its diagnostic process follows an iterative cycle of hypothesis generation, verification, and self-reflection, allowing it to refine conclusions and address logical gaps.

Each diagnosis is supported by a complete chain of evidence, enabling physicians to understand not only the result but also the reasoning behind it, enhancing transparency and trust.

Plans for global collaboration

Sun Kun, one of the study’s corresponding authors from Xinhua Hospital, announced plans to establish a global AI alliance for rare disease diagnosis and treatment. The team aims to complete real-world validation of 20,000 rare disease cases within the next six months.