RareDxR1 AI Model Improves Rare Disease Diagnosis.

Diagnosing rare diseases is a challenging clinical task, requiring physicians to identify precise phenotypes from complex patient symptoms and navigate a vast search space. Existing AI approaches often fall short due to information loss from predefined ontologies, retrieval bottlenecks, or a lack of true diagnostic logic. Researchers have developed RareDxR1, an end-to-end large language model specifically designed for autonomous medical reasoning in open-domain rare disease diagnosis, working directly from unstructured clinical notes. The model employs a progressive training framework that synergizes knowledge internalization with autonomous evolutionary learning, eliminating the need for structured phenotypes or extensive human annotation. It uses Reflection-Enhanced Reasoning Sampling (RERS) to synthesize expert-level diagnostic trajectories by learning from failures, and a dual-level curriculum reinforcement learning approach to master diagnosis. RareDxR1 demonstrates state-of-the-art accuracy across benchmarks, marking a significant advancement in this critical medical field.