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). It processes diverse clinical input data to rank-order diagnostic hypotheses for rare diseases and transparently displays the reasoning process for each hypothesis. It consists of a central host with a long-term memory module and a specialized agent server that integrates over 40 specialized tools and cutting-edge medical knowledge sources. Its modular and scalable design enables it to perform complex diagnostic inference while maintaining traceability and adaptability. Evaluation results using eight datasets demonstrated 100% accuracy for 1,013 of 2,919 diseases, outperforming 15 other methods (existing bioinformatics diagnostic tools, LLMs, and other agent systems). Notably, its Recall@1 score averaged 57.18%, 23.79 percentage points higher than the second-best method (Reasoning LLM). In a multimodal input scenario, the Recall@1 score was 70.60%, which was higher than that of Exomiser (53.20%), and manual validation of the inference process by clinical experts showed a 95.40% agreement rate. It was implemented as a user-friendly web application (http://raredx.cn/doctor) .
Takeaways, Limitations
•
Takeaways:
◦
Demonstrating excellent performance of LLM-based rare disease diagnosis system.
◦
Achieved improved accuracy and Recall@1 scores compared to existing methods.
◦
Possibility of processing diverse clinical data (multimodal).
◦
Provides a transparent and traceable diagnostic reasoning process.
◦
Implemented as a user-friendly web application.
•
Limitations:
◦
Further review of the size and diversity of the evaluation dataset is needed.
◦
Further research is needed to determine generalizability to real-world clinical settings.
◦
Further research is needed on error analysis and improvement measures.
◦
The need for continuous improvement in the accessibility and usability of web applications.
