Automated ECG quality assessment

This repository conains code to assess the noise level present in ECG recordings. ECG recordings can get a binary label of either acceptable or unacceptable. Where acceptable means that the ECG recording contains little noise and is suitable to be used for analysis and diagnosis. And unacceptable means that the ECG recording contains too much noise and is unsuitable for analysis.

The code is divided in two folders. The toolbox and CNN-LSTM:

Toolbox

This folder contains the code for a toolbox to display ECG recordings. Running the toolbox/main.py file will display the following toolbox:

Clinking the Import Data button will open the file explorer where you can select an ECG recording to analyse. This data should be 10 seconds, 500 Hz and between 1 and 12 leads (data shape: nr_leads x 5000). The toolbox is currently able to accept files with the following extensions: .txt, .csv, .hea, .xls, .xlsx, and .wav.. Clicking the Process button will display a results table with the results of individual quality assessment algorithms and the final consensus.

CNN-LSTM

The CNN part of the algorithm consists of three blocks that downsample the data through convolutions. Each block consists of a CNN layer, a max pooling layer, and a dropout layer.

saved_model

This folder contains trained model parameters that can ben loaded into the notebook files instead of training the model again.

Running Toolbox and CNN-LSTM

Downloading and running Toolbox

The toolbox requires the CNN-LSTM model to run. A saved model in the CNN_LSTM/saved_model folder can be used. Alternatively, you can run and save the model yourself. How to do this is explained in the next section.

The path to the model needs to be changed in the parameters_gui file. The requirements_toolbox.txt file in the toolbox folder contains all the packages needed to run the toolbox. $ pip install -r requirements_toolbox.txt should download all the packages needed to run the toolbox in your virtual environment.

Running the toolbox/main.py file will display the toolbox.

Downloading and running CNN-LSTM

The CNN-LSTM runs in python 3.8.11 The requirements_cnn_lstm.txt file contains all the packages needed to run the CNN-LSTM notebooks. $ pip install -r requirements_cnn_lstm.txt should download all the packages needed to run the CNN-LSTM notebooks.

The CinC11 dataset used is the relabeled version by Kramer

There are three versions of the CNN-LSTM notebook with different datasets:

CinC11 and CinC17 unbalanced

Uses the following files: parameters_CINC11CINC17_unbalanced.py, generate_CINC11_CINC17_dataset.ipynb, generate_BUTQDB_dataset.ipynb, data_storage_utils.py, data_preprocessing_utils.py, CNN_LSTM_train_CINC11CINC17_unbalanced_test_butqdb.ipynb

1. Change the path variable names in the parameters_CINC11CINC17_unbalanced.py file.
2. Generate the datasets with generate_CINC11_CINC17_dataset.ipynb in this file change the parameters import to from parameters_CINC11CINC17_unbalanced import parameters. And generate the test dataset with generate_BUTQDB_dataset.ipynb.
3. Run CNN_LSTM_train_CINC11CINC17_unbalanced_test_butqdb.ipynb

CinC11 and CinC17 balanced

Uses the following files: parameters_CINC11CINC17_balanced.py, generate_CINC11_CINC17_dataset.ipynb, generate_BUTQDB_dataset.ipynb, data_storage_utils.py, data_preprocessing_utils.py, CNN_LSTM_train_CINC11CINC17_balanced_test_butqdb.ipynb

1. Change the path variable names in the parameters_CINC11CINC17_balanced.py file.
2. Generate the datasets with generate_CINC11_CINC17_dataset.ipynb in this file change the parameters import to from parameters_CINC11CINC17_balanced import parameters. And generate the test dataset with generate_BUTQDB_dataset.ipynb.
3. Run CNN_LSTM_train_CINC11CINC17_balanced_test_butqdb.ipynb

Kramer

Uses the following files: parameters_kramer.py, generate_kramer_dataset_train_and_test.ipynb, data_storage_utils.py, data_preprocessing_utils.py, CNN_LSTM_train_kramer_test_kramer.ipynb

1. Change the path variable names in the parameters_kramer.py file.
2. Generate the datasets with generate_kramer_dataset_train_and_test.ipynb in this file change the parameters import to from parameters_kramer import parameters.
3. Run CNN_LSTM_train_kramer_test_kramer.ipynb