Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
DeepRare, an artificial AI system integrating 40 specialized tools, outperformed medical specialists in identifying rare conditions in a head-to-head study published in Nature.
For millions of people with rare diseases, the road to diagnosis is a maze. Patients fall between general practitioners and specialists over the years, sometimes decades, combining symptoms that fall outside of textbook presentations.
Eighty percent of rare diseases have a genetic origin, but most are not diagnosed until too much biological damage has occurred. The bottleneck is not a lack of data, it is finding the needle in the medical haystack.
A new study published in Nature this month suggests that artificial intelligence could speed up that hunt. Researchers at the School of Artificial Intelligence of Shanghai Jiao Tong University and Xinhua Hospital developed DeepRarean AI system designed to mimic how human doctors reason through diagnostic uncertainty.
In a head-to-head comparison with five experienced doctors, each with more than ten years of practice, the system achieved higher accuracy across the board.
The numbers are impressive. DeepRare correctly identified the disease on its first suggestion 64.4 percent of the time, compared to 54.6 percent for doctors. When given three suggestions instead of one, the AI system achieved diagnostic success in 79 percent of cases versus 66 percent for human specialists.
Crucially, doctors endorsed the AI’s reasoning 95.4 percent of the time, suggesting that the system not only reaches correct conclusions, but does so in ways that experienced clinicians find persuasive and medically sound.
What sets DeepRare apart from previous diagnostic AI is its architecture. Instead of applying a black-box classification model, the system integrates 40 specialized digital tools and follows an explicitly reasoned workflow.
He forms diagnostic hypotheses, tests them against patient evidence, searches global medical literature databases, analyzes genetic variants, and iteratively reviews his conclusions before ranking the possibilities.
The process mirrors the cognitive steps a human diagnostician takes, but with access to all the medical knowledge and computing speed humans can’t match.
The system has already moved beyond the laboratory. As of July 2025, DeepRare has been deployed on an online diagnostic platform, with more than 600 medical institutions worldwide registered to use it.
The research team plans to validate the system further using 20,000 cases in the real world and to launch a global diagnostic alliance of rare diseases. Notably, the authors emphasize that the system is not intended to replace clinics, but to augment diagnostic workflows, a safeguard that recognizes both the technical limits of AI and the irreducible human element in medicine.
The implications for patients are profound. About 300 million people worldwide are affected by rare diseases, and the average diagnostic odyssey spans five years or more.
Every year of diagnostic delay is a year of uncertainty, wrong treatments, and organ damage that accumulates. An AI system that can shave weeks or months off that timeline, and surface possibilities that might otherwise be overlooked, could reshape the early life experience with a rare condition.
