[China Tech] Chinese Experts Develop World's First AI System With Traceable Reasoning for Rare Disease Diagnosis

China Tech is a column dedicated to the innovations reshaping China – and, inevitably, the world. From cutting-edge AI labs and next-gen robotics to homegrown apps that redefine daily life, we explore the breakthroughs that emerge from the country's relentless drive for technological dominance. Some are game-changers, others cautionary tales, but all offer a glimpse into the future as it's being built, at breakneck speed, in China.

In a groundbreaking development for global healthcare, Chinese researchers have unveiled the world's first artificial intelligence-assisted rare disease diagnosis system featuring "traceable reasoning." The innovative system, named DeepRare, represents a significant leap forward in addressing the longstanding challenges of rare disease diagnosis, with its findings now published in the prestigious journal Nature.

Breaking the 'Black Box' Barrier

One of the most persistent challenges in applying AI to medical diagnosis has been the "black box" problem – traditional AI models could provide results but failed to explain how they arrived at their conclusions, creating understandable trust issues among medical professionals.

DeepRare overcomes this limitation through its pioneering use of an agentic AI architecture. Unlike conventional systems that merely retrieve information, DeepRare demonstrates a deep comprehension of medical knowledge by integrating real-time medical literature with actual clinical data.

What sets this system apart is its human-like diagnostic thinking process. It mimics the cognitive approach of experienced physicians by asking questions, forming hypotheses, verifying them against available data, and reflecting on potential loopholes. Most importantly, it provides doctors with a complete evidence chain for every diagnostic result, ensuring transparency and building trust in AI-assisted decision-making.

Clinical Excellence Through Data

The system's effectiveness has been validated through rigorous testing, demonstrating remarkable accuracy even in challenging circumstances:

· Without genetic data: DeepRare achieves 57.18 percent diagnostic accuracy – an impressive 23.79 percentage points higher than the world's best existing AI models. This capability is particularly valuable for grassroots hospitals that may lack advanced genetic testing equipment.

· With genetic data integration: For complex cases, accuracy soars to 70.6 percent, significantly outperforming the current mainstream model Exomiser, which achieves 53.2 percent accuracy.

Bridging the Healthcare Gap

Dr Sun Kun from Xinhua Hospital, a leading expert in the research, highlighted the system's transformative potential for primary healthcare settings. "Rare disease diagnosis traditionally relies heavily on genetic testing," Sun explained. "For grassroots hospitals, the absence of genetic testing devices and specialized expertise makes rare disease diagnosis exceptionally difficult. The development of DeepRare helps solve these fundamental challenges."

The system effectively serves as a "key" for primary screening at basic medical institutions, democratizing access to rare disease diagnostic capabilities. For complicated cases requiring deeper analysis, the integration of genetic data enables the system to deliver highly accurate diagnoses that rival specialized medical centers.

Global Recognition and Adoption

Since its launch in July last year, DeepRare has garnered significant international attention. The system has already been adopted by over 600 medical facilities and research institutions worldwide, with more than 1,000 medical professionals registering to implement the technology.

This widespread adoption reflects the global medical community's recognition of DeepRare's potential to revolutionize rare disease diagnosis. By combining cutting-edge AI technology with transparent, evidence-based reasoning, the system offers new hope for millions of rare disease patients worldwide who have long struggled with misdiagnosis and delayed treatment.

The publication of DeepRare's development in Nature underscores its scientific significance and positions it as a milestone in the application of artificial intelligence to precision medicine, potentially paving the way for a new generation of transparent, trustworthy AI diagnostic systems, local experts said.