RareDxR1: Autonomous Medical Reasoning for Rare Disease Diagnosis Beyond Human Annotation

Short summary

RareDxR1 is a novel LLM that diagnoses rare diseases directly from unstructured clinical notes, bypassing phenotype extraction bottlenecks. It learns expert diagnostic reasoning through Reflection-Enhanced Reasoning Sampling without human annotation, enhanced by curriculum reinforcement learning. Achieves state-of-the-art accuracy; code and dataset to be released.

• •Novel LLM for end-to-end rare disease diagnosis from unstructured clinical notes
• •Learns expert diagnostic reasoning via Reflection-Enhanced Reasoning Sampling without human annotation
• •Achieves state-of-the-art accuracy with curriculum reinforcement learning; code to be open-sourced

Generated with AI, which can make mistakes.