CTPelvic1K Dataset

Dataset Description

The CTPelvic1K dataset for pelvic bone segmentation from CT scans. This dataset contains CT scans with dense segmentation annotations.

Dataset Details

• Modality: CT
• Target: sacrum, left hip, right hip, lumbar vertebrae
• Format: NIfTI (.nii.gz)

Dataset Structure

Each sample in the JSONL file contains:

{
  "image": "path/to/image.nii.gz",
  "mask": "path/to/mask.nii.gz",
  "label": ["organ1", "organ2", ...],
  "modality": "CT",
  "dataset": "CTPelvic1K",
  "official_split": "train",
  "patient_id": "patient_id"
}

Usage

Load Metadata

from datasets import load_dataset

# Load the dataset
ds = load_dataset("MedOtter/CTPelvic1K")

# Access a sample
sample = ds['train'][0]
print(f"Patient ID: {sample['patient_id']}")
print(f"Image: {sample['image']}")
print(f"Mask: {sample['mask']}")
print(f"Labels: {sample['label']}")

Load Images

from huggingface_hub import snapshot_download
import nibabel as nib
import os

# Download the full dataset
local_path = snapshot_download(
    repo_id="MedOtter/CTPelvic1K",
    repo_type="dataset"
)

# Load a sample
sample = ds['train'][0]
image = nib.load(os.path.join(local_path, sample['image']))
mask = nib.load(os.path.join(local_path, sample['mask']))

# Get numpy arrays
image_data = image.get_fdata()
mask_data = mask.get_fdata()

print(f"Image shape: {image_data.shape}")
print(f"Mask shape: {mask_data.shape}")

Citation

@article{ctpelvic1k,
  title={CTPelvic1K: A Large-Scale Dataset for Pelvic Bone Segmentation},
  year={2023}
}

Dataset Homepage

https://zenodo.org/record/4588403

License

CC BY 4.0 (Zenodo 4588403: annotations, the CLINIC and CLINIC-metal images, models); the mirror carries only dataset6 CLINIC and dataset7 CLINIC-metal

Redistribution and commercial use are permitted under the terms below.

Source of the terms: https://zenodo.org/records/4588403

Audit note (verbatim from the MedOtter dataset card):

2026-09-06 adjudication (settles the 2026-08-17 conditional): Zenodo record 4588403 (licence cc-by-4.0, access_right open) hosts 'Annotations and new collected clinical data in CTPelvic1K dataset': for datasets 1-5 only the annotations (*_mask_mappingback.tar.gz), for dataset6 (CLINIC) and dataset7 (CLINIC-metal) the images too. The mirror carries exactly dataset6 and dataset7 with their masks (103 + 14 cases, verified 2026-09-06) and nothing from datasets 1-5, whose images live under their own upstream licences (KiTS19 CC BY-NC-SA 4.0, BTCV Abdomen/Cervix no licence text, COLONOG CC BY 3.0, MSD CC BY-SA 4.0). Every byte the mirror hosts is therefore within the Zenodo CC BY 4.0 grant: redistribution yes, commercial use yes, attribution required. Adding any dataset1-5 images to the mirror re-opens this verdict (the BTCV subsets would make it no). Supersedes the 2026-08-17 conditional, which judged the whole seven-source dataset.

Please cite:

Pengbo Liu, Hu Han, Yuanqi Du, Heqin Zhu, Yinhao Li, Feng Gu, Honghu Xiao, Jun Li, Chunpeng Zhao, Li Xiao, Xinbao Wu, S. Kevin Zhou. "Deep Learning to Segment Pelvic Bones: Large-scale CT Datasets and Baseline Models." International Journal of Computer Assisted Radiology and Surgery (2021). https://doi.org/10.1007/s11548-021-02363-8