AI System Boosts Rare Disease Diagnosis Accuracy by 24%

A recent artificial intelligence (AI) system, dubbed DeepRare, has demonstrated a significant improvement in the accuracy of diagnosing rare diseases, achieving a 24 percentage point increase compared to existing models. Developed by a collaborative research team from Xinhua Hospital affiliated with Shanghai Jiaotong University School of Medicine and Shanghai Jiaotong University, DeepRare is now being utilized by over 600 medical institutions globally.

Understanding DeepRare’s Diagnostic Approach

DeepRare distinguishes itself from previous medical AI systems by its ability to track the reasoning process behind its diagnoses. Unlike AI that simply provides a conclusion, DeepRare employs an “evidence chain” inference model, mirroring the step-by-step diagnostic approach of an experienced physician. This transparency builds trust and allows clinicians to understand the basis for the AI’s recommendations.

The Evidence Chain Process

The DeepRare system operates through four key stages:

• Hypothesis Generation: When presented with a patient’s clinical symptoms, DeepRare generates a list of potential rare diseases that could be responsible.
• Evidence Search: The system then searches global medical databases and rare disease case data to compare the patient’s symptoms with each candidate disease.
• Inference Chain Construction: DeepRare constructs a logical framework to determine which symptoms align with each disease candidate and identifies necessary additional tests.
• Self-Correction Loop: The system incorporates new information, refines its hypotheses, and prioritizes potential diagnoses to arrive at the most accurate conclusion.

Improved Diagnostic Accuracy

Initial testing of DeepRare, based on patient clinical symptoms alone, yielded a diagnostic accuracy of 57.18%. This represents a substantial improvement over the performance of existing top-tier international models. When genetic data is integrated, the diagnostic accuracy surpasses 70%. This enhanced accuracy is particularly valuable for medical institutions lacking genetic testing capabilities, enabling them to identify potential rare disease cases more effectively.

Global Implementation and Future Plans

DeepRare has been operational through an online diagnosis platform since July of last year and has quickly gained traction, with over 600 medical institutions worldwide currently utilizing the system. Its reach extends beyond China, with research institutes in Europe and the United States also employing DeepRare in their diagnostic processes. Shanghai Xinhua Hospital is preparing for a formal launch to assist medical staff in reducing diagnostic errors related to rare diseases.

The research team plans to further validate the system through analysis of 20,000 real-world cases over the next six months and intends to establish a global rare disease treatment alliance to facilitate collaboration and improve patient care.