realpulmo/realpulmo
ReaPulmo Dataset Overview ReaPulmo (Real-world Pulmonary) is a high-quality clinical reasoning benchmark consisting of 188 complex respiratory cases. It is designed to test the ability of Large Language Models (LLMs) to perform iterative diagnostic reasoning as new clinical evidence is revealed. Data Schema The dataset follows a structured, multi-stage format for each case: Patient Information & Chief Complaint: Baseline demographics and initial… See the full description on the dataset page: https://huggingface.co/datasets/realpulmo/realpulmo.
ReaPulmo Dataset
Overview
ReaPulmo (Real-world Pulmonary) is a high-quality clinical reasoning benchmark consisting of 188 complex respiratory cases. It is designed to test the ability of Large Language Models (LLMs) to perform iterative diagnostic reasoning as new clinical evidence is revealed.
Data Schema
The dataset follows a structured, multi-stage format for each case:
- Patient Information & Chief Complaint: Baseline demographics and initial symptoms.
- Initial Differential Diagnoses: Expert-curated list of potential diseases with supporting/opposing evidence.
- Recommended Exams: Tests a clinician would order at the initial stage.
- Exam Results: Detailed laboratory and imaging findings (including image URLs).
- Updated Differential Diagnoses: Re-evaluation of the case based on new data.
- Final Diagnosis: The confirmed gold-standard diagnosis.
- Treatment Plan: Clinical management and the patient's response to therapy.
Example Fields
For a specific case (e.g., case_1), the data includes:
- Case ID:
case_1 - Age/Sex: 21-year-old Male
- Findings: Diffuse cystic lesions, thickening of the pituitary stalk.
- Gold Diagnosis: Langerhans cell histiocytosis (LCH).
Usage for NeurIPS
This dataset provides a Croissant-compliant metadata file. You can access the full JSON-LD by using the Hugging Face Dataset Metadata API or by clicking the "Export Croissant" button on the dataset home page.
Disclaimer
Warning: ReaPulmo is for research and benchmarking purposes only. It should not be used to guide real-world medical decisions or for autonomous patient diagnosis.
