This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
DeepRare is a rare disease diagnosis agent system based on a large-scale language model (LLM), which processes diverse clinical inputs and ranks diagnostic hypotheses for rare diseases. Each hypothesis is accompanied by a transparent inference process that connects verifiable medical evidence and intermediate analysis steps. Its modular design, including a specialized agent server and a long-term memory module that integrates more than 40 specialized tools and up-to-date medical knowledge sources, performs complex diagnostic inference while maintaining traceability and adaptability. In the evaluation results using eight datasets, it achieved 100% accuracy for 1,013 out of 2,919 diseases, outperforming 15 methods including existing bioinformatics diagnostic tools, LLM, and other agent systems. In particular, the Recall@1 score was 57.18% on average, which was 23.79%p higher than the second best method (Reasoning LLM). In the multi-modal input scenario, the Recall@1 score was 70.60%, surpassing Exomiser's 53.20%. Manual validation of the inference process by clinical experts showed a 95.40% agreement rate and was implemented as a user-friendly web application (http://raredx.cn/doctor) .
Takeaways, Limitations
•
Takeaways:
◦
Shows potential to significantly improve the accuracy and efficiency of rare disease diagnosis.
◦
Demonstrate the effectiveness of a modular design that integrates diverse clinical data to perform diagnostic inference.
◦
We demonstrate the usefulness of LLM-based agent systems through superior performance compared to existing methods.
◦
Increased applicability to real-world medical settings through user-friendly web applications.
•
Limitations:
◦
Since it was evaluated based on a limited dataset, additional validation on various environments and data is needed.
◦
Although transparency of the reasoning process has been achieved, further research is needed to ensure a complete understanding and reliability of the reasoning process.
◦
Additional improvements may be needed to improve the accessibility and usability of your web application.
◦
Consideration should be given to the potential for bias or error due to the limitations of the LLM.
