Awesome

MSTGCN

Multi-View Spatial-Temporal Graph Convolutional Networks with Domain Generalization for Sleep Stage Classification. [paper]

This work is an extension of previous work: GraphSleepNet: Adaptive Spatial-Temporal Graph Convolutional Networks for Sleep Stage Classification (IJCAI 2020) [paper].

model_architecture

These are source code and experimental setup for the ISRUC-S3 dataset.

Citation

If you find this useful, please cite our work as follows:

@ARTICLE{9530406,
  author={Jia, Ziyu and Lin, Youfang and Wang, Jing and Ning, Xiaojun and He, Yuanlai and Zhou, Ronghao and Zhou, Yuhan and Lehman, Li-wei H.},
  journal={IEEE Transactions on Neural Systems and Rehabilitation Engineering}, 
  title={Multi-View Spatial-Temporal Graph Convolutional Networks With Domain Generalization for Sleep Stage Classification}, 
  year={2021},
  volume={29},
  number={},
  pages={1977-1986},
  doi={10.1109/TNSRE.2021.3110665}}

Datasets

We evaluate our model on the ISRUC-Sleep-S3 dataset and the Montreal Archive of Sleep Studies (MASS)-SS3 dataset.

The ISRUC-Sleep-S3 dataset is available here, and we provide the pipeline to run MSTGCN on it.
The MASS-SS3 dataset is an open-access and collaborative database of laboratory-based polysomnography (PSG) recordings. Information on how to obtain it can be found here.

Requirements

Python 3.6
CUDA 10.1
cuDNN 7.6.5
Tensorflow-gpu 1.15.0
Keras 2.3.1
numpy 1.15.4
scipy 1.1.0
scikit-learn 0.21.3

How to run

1. Get Dataset:

You can download ISRUC-Sleep-S3 dataset by the following command, which will automatically download the raw data and extracted data to ./data/ISRUC_S3/:
```
./get_ISRUC_S3.sh
```
For MASS-SS3 dataset, you need to you need to visit the link and contact their administrator.
2. Data preparation:

To facilitate reading, we preprocess the dataset into a single .npz file:
```
python preprocess.py
```
In addition, distance based adjacency matrix is provided at ./data/ISRUC_S3/DistanceMatrix.npy.
3. Configuration:

Write the config file in the format of the example.

We provide a config file at /config/ISRUC.config
4. Feature extraction:

Run python train_FeatureNet.py with -c and -g parameters. After this step, the features learned by a feature net will be stored.
- -c: The configuration file.
- -g: The number of the GPU to use. E.g.,0,1,3. Set this to-1 if only CPU is used.
```
python train_FeatureNet.py -c ./config/ISRUC.config -g 0
```
5. Train MSTGCN:

Run python train_MSTGCN.py with -c and -g parameters. This step uses the extracted features directly.
```
python train_MSTGCN.py -c ./config/ISRUC.config -g 0
```
6. Evaluate MSTGCN:

Run python evaluate_MSTGCN.py with -c and -g parameters.
```
python evaluate_MSTGCN.py -c ./config/ISRUC.config -g 0
```

Summary of commands to run:

./get_ISRUC_S3.sh
python preprocess.py
python train_FeatureNet.py -c ./config/ISRUC.config -g 0
python train_MSTGCN.py -c ./config/ISRUC.config -g 0
python evaluate_MSTGCN.py -c ./config/ISRUC.config -g 0