---
title: Pelvic Bone Fragments with Injuries Segmentation Challenge
description: Pelvic fracture segmentation challenge for CT scans and synthetic X-ray images
authors:
- name: Minh Toan Dinh
email: mtoan65@proton.me
github: mtoan65
orcid: 0009-0000-3405-4638
version: 1.0.0
license: cc-by-4.0
dataset:
- name: PENGWIN Task 1
description: Pelvic Fracture Segmentation on CT
doi: 10.5281/zenodo.10927452
- name: PENGWIN Task 2
description: Pelvic Fragment Segmentation on Synthetic X-ray Images
doi: 10.5281/zenodo.10913196
tags:
- medical-imaging
- segmentation
- pelvic-fracture
- CT
- X-ray
---
# ðĶī Pelvic Bone Fragments with Injuries Segmentation Challenge ðĨ

## ð Welcome to the [PENGWIN segmentation challenge](https://pengwin.grand-challenge.org/)!
Pelvic fractures, typically resulting from high-energy traumas, are among the most severe injuries, characterized by a disability rate over 50% and a mortality rate over 13%, ranking them as the deadliest of all compound fractures. The complexity of pelvic anatomy, along with surrounding soft tissues, makes surgical interventions especially challenging. Recent years have seen a shift towards the use of robotic-assisted closed fracture reduction surgeries, which have shown improved surgical outcomes. Accurate segmentation of pelvic fractures is essential, serving as a critical step in trauma diagnosis and image-guided surgery. In 3D CT scans, fracture segmentation is crucial for fracture typing, pre-operative planning for fracture reduction, and screw fixation planning. For 2D X-ray images, segmentation plays a vital role in transferring the surgical plan to the operating room via registration, a key step for precise surgical navigation.
## ð Challenge Overview
As a MICCAI 2024 challenge, the PENGWIN segmentation challenge is designed to advance the development of automated pelvic fracture segmentation techniques in both 3D CT scans (Task 1) and 2D X-ray images (Task 2), aiming to enhance their accuarcy and robustness. Our dataset comprises CT scans from 150 patients scheduled for pelvic reduction surgery, collected from multiple institutions using a variety of scanning equipment. This dataset represents a diverse range of patient cohorts and fracture types. Ground-truth segmentations for sacrum and hipbone fragments have been semi-automatically annotated and subsequently validated by medical experts. Furthermore, we have generated high-quality, realistic X-ray images and corresponding 2D labels from the CT data using the DeepDRR method, incorporating a range of virtual C-arm camera positions and surgical tools.
The PENGWIN segmentation challenge consists of two main tasks:
1. **[Task 1: Pelvic fragment segmentation on 3D CT](./Raw/Task_01/README.MD)**
- Segment pelvic fractures in 3D CT scans
- Dataset: 150 CT scans from diverse patient cohorts
2. **[Task 2: Pelvic fragment segmentation on 2D X-ray](./Raw/Task_02/README.MD)**
- Segment pelvic fragments in 2D synthetic X-ray images
- Dataset: 50,000 synthetic X-ray images derived from 100 CT scans
## ðïļ Repository Structure
This repository is organized as follows:
```
./
âââ assets/
â âââ PENGWIN_banner_vp9y9n3.x10.jpeg
â âââ task_1.1.jpg
â âââ task_1.2.jpg
â âââ task_2.1.png
â âââ task_2.2.png
âââ Raw/
â âââ Task_01/ # Task 1 dataset and utilities
â â âââ PENGWIN_CT_train_images_part1.zip
â â âââ PENGWIN_CT_train_images_part2.zip
â â âââ PENGWIN_CT_train_labels.zip
â â âââ README.MD # Detailed information about Task 1
â âââ Task_02/ # Task 2 dataset and utilities
â âââ archive_subfolders.sh
â âââ pengwin_utils.py
â âââ README.MD # Detailed information about Task 2
â âââ requirements.txt
â âââ train/
â âââ input/
â â âââ images/
â â âââ x-ray/
â â âââ 001-010.tar.gz
â â âââ 011-020.tar.gz
â â âââ ...
â âââ output/
â âââ images/
â âââ x-ray/
â âââ 001-010.tar.gz
â âââ 011-020.tar.gz
â âââ ...
âââ README.md # This file
```
## ð Getting Started
To participate in the PENGWIN challenge ð:
1. **ðĨ Download the dataset** from the provided Zenodo links or follow the steps below:
1. **ð§ Setup Git LFS**:
```sh
curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
sudo apt-get install git-lfs
```
2. **ð Create and navigate to the Dataset directory**:
```sh
mkdir Pelvic_Bone_Fragments_with_Injuries_Segmentation_Challenge
cd ./Pelvic_Bone_Fragments_with_Injuries_Segmentation_Challenge
```
3. **ð Initialize Git and add the repository**:
```sh
git init
git remote add origin https://mtoan65:@huggingface.co/datasets/mtoan65/Pelvic_Bone_Fragments_with_Injuries_Segmentation_Challenge
```
4. **âïļ Install Git LFS hook for the repository**:
```sh
git lfs install
```
5. **âŽïļ Pull the repository**:
```sh
git checkout -b main
git pull origin main
```
2. **ð Choose the task you want to work on** (Task 1, Task 2, or both).
3. **ð Follow the instructions in the respective README files**:
- [Task 1 README](./Raw/Task_01/README.MD)
- [Task 2 README](./Raw/Task_02/README.MD)
4. **ðĶ Install the required dependencies for each task.**
5. **ð Start developing your segmentation algorithms!**
## ð Citation
If you use the PENGWIN datasets or challenge in your research, please cite the following:
For Task 1:
```bibtex
@dataset{sang_yudi_2024_10927452,
author = {Sang, Yudi and
Liu, Yanzhen and
Yibulayimu, Sutuke and
Zhu, Gang and
Wang, Yu and
Killeen, Benjamin and
Liu, Mingxu and
Ku, Ping-Cheng and
Armand, Mehran and
Unberath, Mathias and
Wu, Xinbao and
Zhao, Chunpeng},
title = {{PENGWIN Task 1: Pelvic Fracture Segmentation on
CT}},
month = apr,
year = 2024,
publisher = {Zenodo},
version = {v1},
doi = {10.5281/zenodo.10927452},
url = {https://doi.org/10.5281/zenodo.10927452}
}
```
For Task 2:
```bibtex
@dataset{killeen_benjamin_2024_10913196,
author = {Killeen, Benjamin and
Liu, Mingxu and
Ku, Ping-Cheng and
Yudi, Sang and
Liu, Yanzhen and
Yibulayimu, Sutuke and
Zhu, Gang and
Wu, Xinbao and
Zhao, Chunpeng and
Wang, Yu and
Armand, Mehran and
Unberath, Mathias},
title = {{PENGWIN Task 2: Pelvic Fragment Segmentation on
Synthetic X-ray Images}},
month = apr,
year = 2024,
publisher = {Zenodo},
version = {1.0.0},
doi = {10.5281/zenodo.10913196},
url = {https://doi.org/10.5281/zenodo.10913196}
}
```
## ð Additional Information
For more details about the PENGWIN challenge, please visit the [official challenge webpage](https://grand-challenge.org).
## ð License
The PENGWIN datasets are distributed under the Creative Commons Attribution 4.0 International License.
---
## ðĪ About me
[](https://github.com/mtoan65)
[](mailto:mtoan65@proton.me)
[](https://www.linkedin.com/in/mtoan65/)
[](https://x.com/mtoan65)
[](https://huggingface.co/mtoan65)
[](https://github.com/mtoan65)
[](https://discordapp.com/users/mtoan65#7866)
[](https://orcid.org/0009-0000-3405-4638)