CT-SAM3D

Towards a Comprehensive, Efficient and Promptable Anatomic Structure Segmentation Model using 3D Whole-body CT Scans.

Overview

Installation

git clone https://github.com/henguo/ct-sam3d.git
cd ct-sam3d
pip install -e .
# install the official itkwidgtes to get the 'imjoy-jupyterlab-extension' ready
pip install 'itkwidgets[lab]==1.0a49'
# install the customized interactive tool
pip install git+https://github.com/henguo/itkwidgets.git
pip install numpy==1.22.4

Download Model

Please download our model from 🤖️ModelScope.

Quick Start

Please start your interactive segmentation journey from our Notebook.

Tool Usage

Left-click to add a positive point, and right-click to add a negative point.

Download Dataset

If you are interested in the enhanced TotalSeg++ dataset with 107 anatomical structures, you can also download it from 🤖️ModelScope. Please note that all images and masks have been rotated into RAI orientation, and the abnormal case s0341 has been fixed.

Demo

License

CT-SAM3D is released under the Apache 2.0 license.

Citation

If you find this project useful in your research, please cite the following paper:

@misc{guo2024ctsam3d,
      title={Towards a Comprehensive, Efficient and Promptable Anatomic Structure Segmentation Model using 3D Whole-body CT Scans}, 
      author={Heng Guo and Jianfeng Zhang and Jiaxing Huang and Tony C. W. Mok and Dazhou Guo and Ke Yan and Le Lu and Dakai Jin and Minfeng Xu},
      year={2024},
      eprint={2403.15063},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2403.15063}, 
}

Acknowledgements

This code is based on the implementations of SAM, ResT, SPADE, and itkwidgets. We are profoundly grateful for these exceptional and inspiring works！