DeepRare AI system uses multi-agent approach to diagnose rare diseases

DeepRare AI system aims to diagnose rare diseases

Researchers have developed an AI system called DeepRare to help diagnose rare diseases. The multi-agent system is powered by large language models and integrates over 40 specialized medical tools and knowledge sources.

It is designed to address the critical challenge of timely diagnosis for over 300 million people worldwide affected by rare conditions. Patients often face a diagnostic odyssey lasting more than five years.

How the diagnostic AI system works

DeepRare processes various clinical inputs to generate ranked diagnostic hypotheses. It can handle free-text descriptions, structured human phenotype ontology terms, and genetic testing results.

The system provides transparent reasoning linked to verifiable medical evidence. This approach aims to reduce misdiagnoses and unnecessary interventions that delay treatment.

Key capabilities of the system include:

• Integrating more than 40 specialized tools and updated knowledge sources
• Processing heterogeneous clinical data types
• Generating diagnostic hypotheses with evidence-based reasoning

DeepRare outperforms existing methods in testing

Researchers evaluated DeepRare across nine datasets from literature, case reports, and clinical centers. The testing spanned Asia, North America, and Europe and covered 14 medical specialties and 2,919 diseases.

In human-phenotype-ontology-based tasks, DeepRare achieved an average Recall@1 of 57.18%. This performance outperformed the next best method by 23.79%.

In multi-modal tests, the system reached 69.1% accuracy compared to Exomiser's 55.9% on 168 cases. Expert review achieved 95.4% agreement on the validity and traceability of its reasoning chains.

The broader impact on clinical workflows

The research demonstrates how LLM-driven agentic systems could reshape clinical workflows. The technology addresses the substantial emotional and economic burden caused by diagnostic delays.

By providing decision support for differential diagnosis, such systems may help end the prolonged diagnostic odyssey many rare disease patients endure. The work advances both rare disease diagnosis and the application of AI in medicine.