Medical-CXR-VQA: Visual Question Answering Subsets

This dataset is associated with the paper CCD: Mitigating Hallucinations in Radiology MLLMs via Clinical Contrastive Decoding.

Project Page: https://x-izhang.github.io/CCD/

Code: https://github.com/X-iZhang/CCD

Medical-CXR-VQA aggregates 78K expert-authored question–answer pairs sourced from chest radiology studies. Each entry ties a chest X-ray with a clinically grounded question that probes findings, anatomy, or acquisition details.

📚 Dataset Overview

• Provides a held-out benchmark for radiology-focused VQA assessment.
• Matches the evaluation settings in CCD for cross-dataset comparisons.

---

🧾 Data Format

Every record includes:

• subject_id, study_id, dicom_id: identifiers linking back to the originating exam.
• path: relative image path for the associated chest X-ray.
• question: natural-language clinical query.
• question_type: category label (e.g., diagnosis, modality, yes/no).
• answer: short textual answer that models must predict.

🚀 How to Use

from datasets import load_dataset

ds = load_dataset("X-iZhang/Medical-CXR-VQA", split="test")

sample = ds[0]
print(sample["question"], "->", sample["answer"])
print("type:", sample["question_type"], "dicom:", sample["dicom_id"])

✏️ Citation

@article{zhang2025ccd,
  title={CCD: Mitigating Hallucinations in Radiology MLLMs via Clinical Contrastive Decoding},
  author={Zhang, Xi and Meng, Zaiqiao and Lever, Jake and Ho, Edmond SL},
  journal={arXiv preprint arXiv:2509.23379},
  year={2025}
}