The difficulty in diagnosing rare diseases and the high rate of missed diagnosis are common pain points in the global medical field. On the 19th, it was learned that DeepRare, the world's first traceable intelligent rare disease diagnosis system developed by a Chinese expert team, has solved the global problem of rare disease diagnosis. Recently, the joint research results led by Professor Kun Sun and Professor Yongguo Yu from the Affiliated Xinhua Hospital of Shanghai Jiao Tong University School of Medicine, and Professor Ya Zhang and Associate Professor Weidi Xie from the School of Artificial Intelligence at Shanghai Jiao Tong University, were published online in the internationally renowned academic journal Nature. The journal specially invited Timo Lassmann, an expert from the Children's Health Research Center at the University of Western Australia in Australia, to write and distribute a special article titled "News and Perspectives on AI Successfully Diagnosing Rare Diseases" for review. The comment emphasizes that in the face of major medical decisions, systems like DeepRare that can demonstrate their deduction process can withstand scrutiny from human experts and gain trust, ultimately proving their more practical value, rather than providing a "black box" answer. It is understood that traditional medical AI often falls into a "trust crisis" due to the untraceable reasoning process, just like doctors only provide diagnostic results without explaining the basis for diagnosis and treatment. The Chinese expert team has created the world's first "central avatar" traceable Agentic AI architecture for DeepRare, which surpasses traditional medical AI from three dimensions: knowledge reserve, diagnostic thinking, and reasoning process, providing AI diagnosis with a "clear diagnosis and treatment approach". In terms of knowledge reserves, DeepRare breaks through medical data silos, just like connecting all top medical libraries and clinical case databases around the world, integrating massive medical literature knowledge bases with real clinical case data in real time. It is no longer a simple information retrieval, but a deep understanding and internalization of medical knowledge, mobilizing the world's top medical reserves for every diagnosis. In terms of diagnostic thinking, it breaks away from the traditional AI's "pattern matching fast thinking" and has the "slow thinking" ability of human doctors. It can actively ask questions to supplement patient information, refine diagnostic clues through the iterative cycle of "hypothesis verification self reflection", and correct logical loopholes, just like the professional process of senior doctors repeatedly assessing the condition. In the reasoning process, it implements full process white box reasoning, with each diagnostic conclusion accompanied by a complete chain of evidence, allowing doctors to not only know the diagnostic results, but also have a clearer understanding of the diagnostic basis, completely solving the "trust problem" of AI healthcare. Professor Sun Kun told reporters that the diagnosis and treatment of rare diseases have long faced the industry dilemma of "difficult to diagnose without genetic testing", especially in grassroots hospitals lacking genetic testing conditions, rare disease screening is even more difficult. DeepRare has achieved a significant improvement in the performance of rare disease phenotype screening. Research has shown that DeepRare exhibits strong "phenotype decoding" ability when only providing clinical phenotype information of patients without genetic data. The first accuracy rate of phenotype diagnosis reaches 57.18%, which is 23.79 percentage points higher than the previous international best model. This has created a "golden key" for rapid screening of rare diseases in grassroots hospitals, allowing them to conduct efficient initial screening of rare diseases without relying on complex genetic testing. After introducing gene sequencing data, DeepRare's diagnostic efficiency was further upgraded, with a comprehensive first place diagnostic accuracy rate of over 70.6% in complex cases, significantly better than the internationally recognized Exomiser tool (53.2%). It is reported that the application of DeepRare has taken the lead. The DeepRare Disease Online Diagnosis Platform was officially launched on July 26, 2025, and quickly gained recognition in the global medical research field with its accurate diagnostic capabilities and convenient user experience. In just six months since its launch, the platform has attracted over 1000 professional users to register, covering more than 600 medical and research institutions worldwide. From top tier hospitals in China to well-known laboratories in Europe and America, DeepRare is becoming the "smart stethoscope" for doctors worldwide to diagnose and treat rare diseases. Professor Sun Kun revealed that DeepRare has completed internal deployment and entered the internal testing stage at Xinhua Hospital. It is not only a simple auxiliary diagnostic tool, but also a rigorous "digital quality controller", which will be officially applied to the quality control process of rare disease diagnosis and treatment in the entire hospital. In the complex process of rare disease diagnosis and treatment, it will help doctors identify and fill in gaps, timely discover potential omissions in diagnosis and treatment, maintain the "safety bottom line" of diagnosis, and ensure that every rare disease patient can receive comprehensive and accurate diagnosis and treatment evaluation. In order to benefit more rare disease patients, the joint team of Xinhua Hospital and Shanghai Jiao Tong University is actively preparing to launch the "Global AI Rare Disease Diagnosis and Treatment Alliance" and simultaneously launching the "Ten Thousand Person Clinical Verification Plan". The joint team plans to complete real-world validation of 20000 difficult and rare diseases within the next six months, relying on the global alliance network. This plan is not only to further validate and optimize the performance of the DeepRare algorithm, but also to weave a global intelligent diagnosis network for rare diseases using artificial intelligence technology, allowing medical institutions in various countries to share advanced diagnostic technologies and enabling rare disease patients around the world to be accurately identified and effectively treated. (New Society)