Choroid plexus (ChP) segmentation pipeline for T1-weighted magnetic resonance images

Updated on August 3, 2025. This pipeline was enhanced in the paper published at NeuroImage.

This is the pipeline which proposed in the paper published at Fluids and Barriers of the CNS. In this repository, we offer two methods to execute the ChP segmentation pipeline. One method involves directly executing the Python code, while the other method utilizes Docker.

---

Usage

Input/Output formats

• The following input forms are accepeted:

1. Single NIfTI file (.nii or .nii.gz)
2. folder containing .dcm series
3. folder containing multiple NIfTI files
4. .txt file where each row contains the NIfTI file path

Output

Default path: results/
ChP segmentation is saved in results/cp/3_orig_T1_space/.
You can specify a custom output path.
For detailed intermediate results, jump to Output files structure.

Option 1: Using Docker （Recommend）

Pull the Image

docker pull batjoker1/chp-seg:v2

Run the Container

• CPU Version:

docker run -v $OUTPUT_FOLDER_ON_HOST:/app/results -v $DATA_PATH_ON_HOST:$DATA_PATH_IN_CONTAINER -it --rm batjoker1/chp-seg:v2 bash

• GPU Version (requires NVIDIA driver support):

docker run --gpus all -v $OUTPUT_FOLDER_ON_HOST:/app/results -v $DATA_PATH_ON_HOST:$DATA_PATH_IN_CONTAINER -it --rm batjoker1/chp-seg:v2 bash

Run Inference

Once inside the container, run:

python pipeline.py --input File/Directory

Option 2: Python source code

Environment Requirements

Ensure that TensorFlow >2.4 is installed. Then, install the required packages listed in requirements.txt (no strict version requirements) and Deepbrain (pip install deepbrain --no-deps).

Installation & Setup

Clone the repository and navigate to the project directory:

git clone https://github.com/princeleeee/ChP-Seg.git
cd ChP-Seg

Modify the DeepBrain package to enable TensorFlow 1.x compatibility. This is necessary because DeepBrain was originally implemented with TensorFlow 1.x. See the related issue for more details.

For example, in my case I run the following command to enable TensorFlow 1.x compatibility:

sed -i '1s/.*/import tensorflow.compat.v1 as tf/' /usr/local/lib/python3.8/site-packages/deepbrain/extractor.py # Necessary since Deepbrain is accomplished with Tensorlow 1.x

You can change in the similar way according to your own environment.

Download Model Weights

Download the pre-trained deep learning model weights from Google Drive and place them in the weights folder:

Version	Filename	Description
v1	All_data_trainweights.184-0.96731.h5	ChP model weights
	All_data_trainweights.200-0.05769.h5	LVEN model weights
v2	20241210-220442_all_data_trainbest_weights.h5	ChP model weights (updated)
	All_data_trainweights.200-0.05769.h5	LVEN model weights (unchanged)

mkdir weights

Run Inference

Once everything is set up, you can run the pipeline on a sample input:

python pipeline.py --input demo/I812923.nii.gz  # Example run

Or with updated model saved in 'weights/' folder

python pipeline.py --input demo/I812923.nii.gz --ven_weights weights/All_data_trainweights.200-0.05769.h5 --cp_weights weights/20241210-220442_all_data_trainbest_weights.h5

---

Outputs structure

results/cp/3_orig_T1_space is the path of the ChP segmentation results for input files.

results/
├── file_collections.txt # all files input to the pipeline.
│
├── brain/  # save the results in the preprocessing and skull stripping stage.
│   │
│   ├── 0_resample/ # 1mm^3 RAS+ reorientation and resampled images save path.
│   │   ├── crop_range.txt
│   │   └── xxx.nii.gz
│   │
│   ├── 1_check/ # 3 plane .png check the crop option on 0_resample
│   ├── 1_err/ # 3 plane .png check the crop option on 0_resample
│   ├── 1_img/ # cropped skull stripped images from 0_resample, size: 160*200*160, crop range records is brain/0_resample/crop_range.txt
│   ├── 1_mask/ # brain mask on 1_img
│   │
│   └── 2_resample_inverse/ # restore images in 1_img to original image space.
│
├── ventricle/  # ventricle segmentation results folder
│   │
│   ├── 0_mask/ # segmentations resluts of lateral ventricles, size: 160*200*160
│   │
│   ├── 1_img_crop/ # crop brain/1_img to size 96*96*80
│   ├── 1_mask_crop/ # crop ventricle/0_mask to size 96*96*80
│   │
│   ├── 2_resampledT1_space/ # restore ventricle segmentation mask that matches brain/0_resample
│   │
│   ├── 3_orig_T1_space/ # restore ventricle segmentation mask that matches brain/2_resample_inverse
│   │
│   └── crop_range.txt  # crop range records that generated ventricle/1_img_crop and ventricle/1_mask_crop
│
└── cp/ # choroid plexus segmentation resluts folder
    │
    ├── 0_mask/ # segmentation results of choroid plexus, size 96*96*80
    │
    ├── 1_mask_refine/ # refined segmentaiton results of cp/0_mask
    │
    ├── 2_resampledT1_space/ # ChP segmentation match images in brain/0_resample
    │
    └── 3_orig_T1_space/ # ChP segmentation match images in brain/2_resample_inverse

Citation

If you find our work helpful, please consider citing:

@article{li2024associations,
  title={Associations between the choroid plexus and tau in Alzheimer’s disease using an active learning segmentation pipeline},
  author={Li, Jiaxin and Hu, Yueqin and Xu, Yunzhi and Feng, Xue and Meyer, Craig H and Dai, Weiying and Zhao, Li and Alzheimer’s Disease Neuroimaging Initiative},
  journal={Fluids and Barriers of the CNS},
  volume={21},
  number={1},
  pages={56},
  year={2024},
  publisher={Springer}
}

@article{LI2025121392,
title = {Morphological changes of the choroid plexus in the lateral ventricle across the lifespan: 5551 subjects from fetus to elderly},
journal = {NeuroImage},
volume = {318},
pages = {121392},
year = {2025},
issn = {1053-8119},
doi = {https://doi.org/10.1016/j.neuroimage.2025.121392},
url = {https://www.sciencedirect.com/science/article/pii/S1053811925003957},
author = {Jiaxin Li and Yuxuan Gao and Yunzhi Xu and Weiying Dai and Yueqin Hu and Xue Feng and Dan Wu and Li Zhao}
}