A research team in Shanghai has developed an artificial intelligence system designed to shorten the diagnosis of rare diseases, according to a study published Thursday in the journal “Nature.”

The system, known as DeepRare, was created by researchers at Xinhua Hospital, affiliated with Shanghai Jiao Tong University School of Medicine, and the university’s School of Artificial Intelligence.

Rare diseases affect an estimated 300 million people worldwide across more than 7,000 identified conditions. In China, more than 20 million people are believed to live with rare diseases, and patients often wait years for a confirmed diagnosis.

According to the study, DeepRare analyzes patients’ clinical symptoms and compares them with global medical databases, generating step-by-step diagnostic reasoning rather than a single output. The system is designed to make its inference process traceable, allowing physicians to review how conclusions are reached.

In a News & Views commentary published alongside the study, Timo Lassmann of the University of Western Australia wrote that the system addresses longstanding concerns about the “black box” nature of AI in clinical diagnosis by making its reasoning fully transparent.The researchers said the system is built on what they describe as an “agentic” architecture, combining real-time knowledge retrieval with iterative self-correction. It produces an end-to-end reasoning chain for each case, allowing physicians to review the evidence supporting a diagnosis.

Professor Sun Kun of Xinhua Hospital said diagnosing rare diseases remains particularly challenging in hospitals that lack access to genetic testing, especially at the grassroots level.

In testing, when provided only with patients’ clinical phenotypic information and no genetic data, DeepRare achieved a first-diagnosis accuracy rate of 57.18%, representing a 23.79 percentage-point improvement over a previously leading model, according to the study. The researchers said this suggests the system could support preliminary screening in hospitals without advanced genetic testing capabilities.

The study also reported a 95.4% agreement rate between the system’s evidence assessments and clinical experts’ evaluations.
The online DeepRare platform was launched last July, and has since registered more than 1,000 professional users from over 600 medical and research institutions, according to the team.

According to Sun, the system has been deployed internally at Xinhua Hospital and is undergoing further testing before broader clinical integration.

Over the next six months, researchers plan to launch the Global AI Rare Disease Diagnosis Alliance and validate the system using 20,000 additional rare disease cases.

(Header image: E+/VCG)