In the early hours of February 19th (Beijing Time), the top international academic journal Nature published online the research findings of a joint team from Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine and the School of Artificial Intelligence at Shanghai Jiao Tong University. The team developed the world's first traceable rare disease diagnosis system powered by Agentic AI, titled "DeepRare: An Agentic System for Rare Disease Diagnosis with Traceable Reasoning." Leveraging a pioneering Agentic AI architecture, the system effectively addresses the worldwide challenge of rare disease diagnosis and sets a new world record in diagnostic accuracy. It also marks a significant milestone for China's interdisciplinary biomedical engineering innovations on the global stage.

Diagnostic difficulties and high rates of underdiagnosis of rare diseases have long been major challenges in global healthcare. Traditional medical AI faces a trust crisis due to untraceable reasoning, but DeepRare addresses this issue by achieving end-to-end white-box reasoning, providing diagnostic conclusions backed by a complete chain of evidence. Addressing the industry bottleneck where rare disease diagnosis relies heavily on complex genetic testing, the system achieves a top-1 diagnostic accuracy of 57.18% based solely on clinical phenotypic information, a 23.79 percentage-point improvement over the previous state-of-the-art models. This breakthrough enables primary care hospitals to conduct efficient initial screenings for rare diseases without complex genetic tests.

The online diagnostic platform for DeepRare was launched on July 26th, 2025. Within six months, the platform has registered over 1,000 professional users and provided diagnostic support to more than 600 top medical research institutions worldwide, acting as a "smart stethoscope" for doctors treating rare diseases. Currently, the joint team is preparing to launch the "Global AI Alliance for Rare Disease Diagnosis and Treatment" and initiating the "Ten-Thousand-Patient Clinical Validation Program," aiming to validate 20,000 real-world complex rare disease cases within the next six months. Moving forward, the continuous optimization and global promotion of DeepRare will empower artificial intelligence to play a greater role in rare disease diagnosis and public health security, contributing Chinese expertise and solutions to global healthcare.

中国AI登上Nature！院校联合研发DeepRare，助力罕见病诊断

北京时间2月19日凌晨，国际顶级学术期刊《自然》在线刊发上海交通大学医学院附属新华医院与上海交通大学人工智能学院联合团队研究成果——全球首个可溯源智能体式罕见病诊断系统“DeepRare：An Agentic System for Rare Disease Diagnosis with Traceable Reasoning”。该系统凭借首创的Agentic AI架构破解罕见病诊断世界性难题，诊断精度刷新世界纪录，也标志着中国医工交叉创新成果站上世界舞台中央。

罕见病确诊难、漏诊率高是全球医疗痛点，传统医疗AI因推理不可追溯存在信任危机，而DeepRare实现全流程白盒推理，诊断结论附带完整证据链，彻底解决这一难题。针对罕见病诊疗依赖基因检测的行业困境，该系统仅凭借临床表型信息，表型诊断首位准确率达57.18%，较国际最佳模型提升23.79个百分点，让基层医院无需复杂基因检测即可开展罕见病高效初筛。

这款系统的在线诊断平台已于2025年7月26日上线，半年内吸引超1000名专业用户注册，服务全球600余家顶尖医疗科研机构，成为医生诊疗罕见病的 “智能听诊器”。目前，联合团队正筹备发起“全球AI罕见病诊疗联盟”，同步启动 “万人临床验证计划”，计划半年内完成20000例疑难罕见病真实世界验证。未来，DeepRare的持续优化与全球推广，将让人工智能在罕见病诊疗、公共卫生保障等领域发挥更大作用，为全球医疗健康事业贡献中国智慧与力量。

译｜陈冠臻
英文审校｜柯妃娟