A Comprehensive Rare Disease Diagnostic Dataset with nearly 50,000 patients covering more than 4000 diseases.
For the reproduction and evaluation script, see our Github.
We build our work upon PMC-Patients, a large-scale patient summary dataset sourced from PMC case reports, and we use GPT-4o for all data processing.
To be specific, we first filter cases focusing on rare disease diagnoses from PMC-Patients, and extract their ground-truth… See the full description on the dataset page:
https://huggingface.co/datasets/THUMedInfo/RareArena.
