CheXpert-plus-RRG: Radiology Report Generation Subsets

CheXpert-plus-RRG curates two report-generation benchmarks from the CheXpert Plus validation split. Each configuration links a frontal chest radiograph with its matching clinical narrative.

---

📚 Dataset Overview

Subset	Section Target	Split	#Samples	View Coverage
findings_section	Findings	valid	62	Frontal (AP/PA)
impression_section	Impression	valid	202	Frontal (AP/PA)

• Packaged as a validation-only suite for benchmarking and zero-shot evaluation.
• Mirrors the CCD and Libra evaluation pipeline for comparing report quality across datasets.
• ❗️Further manual screening excluded 10 ambiguous view images (October 1st, 2025).

---

🧾 Data Format

Every sample provides:

• main_image: a frontal-view chest X-ray as a PIL Image object.
• Section text: findings_section or impression_section, selected by the dataset configuration.
• default_prompt: a lightweight prompt scaffold to seed generation models.

---

🚀 How to Use

from datasets import load_dataset

# Load a specific subset (e.g., findings_section)
ds = load_dataset("X-iZhang/CheXpert-plus-RRG", name="findings_section", split="valid")

# Display an image
from PIL import Image
image = ds[0]["main_image"].convert("RGB")
image.show()

# Inspect the paired report text
text_key = "findings_section"  # switch to "impression_section" for that subset
print(ds[0][text_key])

✏️ 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}
}