Ψ-Net: Stacking Densely Convolutional LSTMs for Sub-Cortical Brain Structure Segmentation

by Lihao Liu, Xiaowei Hu, Lei Zhu, Chi-Wing Fu, Jing Qin and Pheng-Ann Heng.

Introduction

In this repository, we provide the Tensorflow and DLTK implementation for our TMI paper Ψ-Net: Stacking Densely Convolutional LSTMs for Sub-Cortical Brain Structure Segmentation.

Requirement

tensorflow-gpu 1.14.0
dltk 0.2.1
cuda 10.0
cudnn 7.5

Usage

Clone the repository:

git clone https://github.com/lihaoliu-cambridge/psi-net.git
cd psi-net

Download the IBSR dataset:

IBSR_V2.0_nifti_stripped.tgz
Unzip them in folder dataset/IBSR:

dataset/IBSR/IBSR_nifti_stripped
Pre-process the LPBA40 dataset:
```
cd script
python preprocessing_ibsr.py
```
output results:

dataset/IBSR_preprocessed/IBSR_nifti_stripped
Train the model:
```
cd ..
python train_ibsr.py
```
Test the saved model:
```
python test_ibsr.py 
```

Note

If you are using a virtual environment, please reload cuda and cudnn before running, so you can use gpu during training. You can also add the cuda and cudnn path to your system path:
```
source ~/tensorflow-env/bin/activate
module load cuda/10.0 cudnn/7.5_cuda-10.0
```

Use pip to install Tensorflow and DLTK directly:

pip install tensorflow-gpu==1.14.0
pip install dltk

In our TMI paper, we use Whiten normalization to standardize data distributions. To better standardize data distributions and facilitate training, we try another normalization approach Histogram Standardization. The results are shown in the below picture:

Citation

If you use our code for your research, please cite our paper:

@article{liu2020psi,
  title={$\psi$-Net: Stacking Densely Convolutional LSTMs for Sub-cortical Brain Structure Segmentation},
  author={Liu, Lihao and Hu, Xiaowei and Zhu, Lei and Fu, Chi-Wing and Qin, Jing and Heng, Pheng-Ann},
  journal={IEEE Transactions on Medical Imaging},
  year={2020},
  publisher={IEEE}
}

Question

Please open an issue or email lhliu1994@gmail.com for any questions.