Gender Classification

This project is to classify the Gender from a 12-Lead ECG Signal.

Author

Felix Tempel

Technologies

Project is created with the Pytorch framework using a 34-Layer ResNet Model. For Optimizing the SGD-Optimizer is used.

Setup

To run this project, clone the repository and install the requirements:

$ git clone 
$ pip install -r requirements.txt

Data

The data for training the model can be found under the following link: https://physionetchallenges.github.io/2020/

For the classification three data sets have been used which are:

CPSC2018 training set, 6.877 recordings

PTB-XL electrocardiography Database, 21.837 recordings

Georgia 12-Lead ECG Challenge Database, 10.344 recordings

The data is sampled at 500Hz and consists of a .hea and .mat file.

The model was trained on the healthy ECG Files.

The data files have to be downloaded into a folder of your desire - the path has to be given as an argument when starting the project

If you have Google Colab account you can run the following command:

./get_data.sh

The folder structure for the data should be the following:

datafolder/
 |
 +-- chin_database
 |  |
 |  +-- files
 |   
 +-- ptb_databse
 |  |
 |  +-- files
 |    
 +-- georgia_database
 |  |  
 |  +-- files
 |    
 +-- mit_database
 |  |  
 |  +-- files
...

Docker

To run this project with Docker build the container in the following way:

git clone
cd proj
docker build -t docker_gender_classification .
docker run --mount type=bind, source="data_root", target="data_root" --gpus all --shm-size=8g -it docker_gender_classification

Terminal

To run the project from the Terminal:

cd gender_classification
python3 main.py -h

positional arguments:
  <command>    train, hypertrain or test
  <database>   Dataframe file location on your system
  <files>      File location on your system

optional arguments:
  -h, --help   show this help message and exit
  --ckpt CKPT  Path to model weights
  --ex EX      Additional save name

Example:

python3 main.py test df_mix_op.csv ./src/datafolder

Visualization

To visualize the model a tensorboard logger is implemented - to view the parameters:

tensorboard --logdir=path/to/tensorboard/file