Segmentations of distended and collapsed colons from CT colonography images
This repository contains the nnU-Net trained to segment the colon from CT colonography images presented in this paper: TBA.
Data used for model training
For details, please have a look at our paper.
- Raw images: CT COLONOGRAPHY (ACRIN 6664) dataset from the The Cancer Imaging Archive
- Masks: Besides masks we obtained from our semi-automatic method, we use masks from the HQColon dataset
Technical setup and energy consumption
The model was trained on a NVIDIA DGX SPARK GB10. We tracked the energy consumption while model training via Carbontracker resulting in a total of 21,94 kWh.
Disclaimer
Please be aware that this model is not intended or desdigned for clinical use. It was developed in a research setting to create so called silver labels and we highly suggest to not just blindly apply this model to a dataset. Instead, be critical and do not consider this as another state-of-the-art model.
Usage
Our setup assumes you have docker installed. We run the following docker container:
docker run -it --gpus=all --shm-size=64g -v <YOUR_DIRECTORY>:/workspace nvcr.io/nvidia/pytorch:26.01-py3
Install dependencies (inside the container):
pip install nnunetv2==2.8.1
Now you are able to run the inference:
# Make sure to modifie the input and output directory inside run_inference.sh
bash run_inference.sh
References
Please have a look at our paper for a more complete list of references.
@misc{anthony2020carbontracker,
title={Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models},
author={Lasse F. Wolff Anthony and Benjamin Kanding and Raghavendra Selvan},
howpublished={ICML Workshop on Challenges in Deploying and monitoring Machine Learning Systems},
month={July},
note={arXiv:2007.03051},
year={2020}
}
@misc{smith2015,
title={Data From CT COLONOGRAPHY (ACRIN 6664) (Version 2) [dataset]. The Cancer Imaging Archive.},
author={Smith, K., Clark, K., Bennett, W., Nolan, T., Kirby, J., Wolfsberger, M., Moulton, J., Vendt, B., Freymann, J},
note={https://doi.org/10.7937/k9/tcia.2015.nwtesay1},
year={2015}
}
@misc{finocchiaro2025hqcolonhybridinteractivemachine,
title={HQColon: A Hybrid Interactive Machine Learning Pipeline for High Quality Colon Labeling and Segmentation},
author={Martina Finocchiaro and Ronja Stern and Abraham George Smith and Jens Petersen and Kenny Erleben and Melanie Ganz},
year={2025},
eprint={2502.21183},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2502.21183},
}
@misc{Isensee2021,
title={nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation.},
author={Isensee, F., Jaeger, P. F., Kohl, S. A., Petersen, J., & Maier-Hein, K. H. },
year={2021},
howpublished{Nature Methods, 18(2), 203-211.}
}