ECG Classification

• This project implements classification models for 2 tasks.

  1. Discrimination of real and synthetic ECGs seprately for Healthy and AF label groups utilizing different classifiers.
  2. Cross-domain classification of Healthy and AF ECGs utilizing real and synthetic ECG features.

• This is a use case for the main project here - https://github.com/dieterich-lab/ecg_data_synthesis

• The main feature extraction pipeline to get the basic ECG features for real and synthetic data is implemented here - https://github.com/dieterich-lab/ecg_processing

Installation

1. Clone the repository

[email protected]:dieterich-lab/ecg_processing.git
cd ecg_processing

2. To install the necessary dependencies, you can use:

pip install -r requirements.txt

Usage

The script supports two main tasks: real vs. synthetic ECG classification and healthy vs. AF classification. You can run the script from the command line with different parameters to perform these tasks. Example commands for both the tasks are shown below and run them from the src directory.

1. Real vs Synthetic Classification: This command will classify real vs synthetic ECG signals for the healthy label, using peak-based features.

python classify.py --task real_vs_synth --label healthy --csv_path path/to/real_synth_healthy.csv --use_peaks True

Parameters:

• --task real_vs_synth: Specifies that the task is to classify real vs synthetic ECG signals.
• --label healthy: Specifies that the classification is for the "healthy" label (you can also use af for atrial fibrillation).
• --use_peaks True: Indicates whether to use peak-based features for classification.

2. Healthy vs AF Classification (with real data): This command will classify healthy vs AF ECGs using real data.

python classify.py --task healthy_vs_af --data real --real_csv_path path/to/real_data.csv --real_labels_path path/to/real_labels.npy

Parameters:

• --task healthy_vs_af: Specifies that the task is to classify healthy vs AF ECGs.
• --data real: Indicates that the data used for classification is real.
• --real_csv_path: The path to the CSV file containing features for real ECG data.
• --real_labels_path: The path to the .npy file containing labels for real ECG data.

3. Healthy vs AF Classification (with synthetic data):

python classify.py --task healthy_vs_af --data synth --synth_csv_path path/to/synth_data.csv --synth_labels_path path/to/synth_labels.npy

Parameters:

• --task healthy_vs_af: Specifies that the task is to classify healthy vs AF ECGs.
• --data synth: Indicates that the data used for classification is synthetic.
• --synth_csv_path: The path to the CSV file containing features for synthetic ECG data.
• --synth_labels_path: The path to the .npy file containing labels for synthetic ECG data.