Automated Sleep Stage Classification

A sequence to sequence learning model for sleep stage and fragmentation analysis with ready to use server API.

Github repo

Requirements

• Python 2.7
• tensorflow/tensorflow-gpu
• numpy
• scipy
• scikit-learn
• matplotlib
• imbalanced-learn(0.4.3)
• pandas
• mne

Data Preparation

We use the Physionet Sleep-EDF datasets for training our model.

• Use the below script to download SC subjects from the Sleep_EDF (2013) dataset

cd data_2013
chmod +x download_physionet.sh
./download_physionet.sh

• Use the below script to preprocess the signals from specified folder, channel

python prepare_upload_signal.py --data_dir data_2013 --output_dir data_2013/eeg_fpz_cz --select_ch 'EEG Fpz-Cz'

Train Model

You can modify args settings in seq2seq_sleep_sleep-EDF.py at your discretion.

• Use the below script to train a model with the 20-fold cross-validation using Fpz-Cz channel of the Sleep_EDF (2013) dataset:

python seq2seq_sleep_sleep_EDF.py --data_dir data_2013/eeg_fpz_cz --output_dir output_2013 --n_folds 20

Generate Classification for a Single Signal

Use the below script to first preprocess a signal and then classify different stages for it with the selected model saved in folder /seq2seq-model.

python prepare_upload_signal.py --data_dir eeg_data --output_dir eeg_data/eeg_fpz_cz --select_ch 'EEG Fpz-Cz'
python predict_upload_signal.py --filename ST7022J0.npz

Sleep Fragmentation Analysis and Scoring

Helper functions for scoring, sleep fragmentation analysis locate in script eeg_frag_info_combine.py. These functions will be used in our API in deploy.py

Deploy as a server for Web APP

Use the below script to start a local Flask server

python deploy.py

Supported API

• POST /api/analysis
  • Body: file=[sleep signal file]
  • Header: ContentType=multipart/form-data
  • Response:
    • message: HTTP status message
    • sleep_score, sleep_msg: sleep quality score and the message to display on clicking the button of score
    • wtr, wtr_msg: weighted transition rate and the message to display on clicking the button of wtr
    • total_sleep_time: total actual sleep time to be displayed on FE
    • first_section_info: the information about the first section
    • total_section_num: total number of sections

    {  
        "message": ...,
        "sleep_score": ...,
        "sleep_msg": ...,
        "wtr": ...,
        "wtr_msg": ...,
        "total_sleep_time": ...,
        "total_section_num": ...,
        "first_section_info": {
            "section_id": ...,
            "stage_name": ...,
            "description": ..., 
            "signal": {
                    "x": ...,
                    "y": ...,
            }
        }
    }
    

• GET /api/section
  • Param: sec_id=[int from 0 to total_section_num-1]
  • Response:

    {
        "section": {
            "section_id": ...,
            "stage_name": ...,
            "description": ..., 
            "signal": {
                    "x": ...,
                    "y": ...,
            }
        }
    }
    

Miscellaneous

• dataloader.py dhedfreader.py are helper libraries for reading edf signal files

• Procfile requirements.txt runtime.txt are config files for hosting the server on Heroku

References

github:akaraspt
github:MousaviSajad