ECG Multi-Label Classification

This repository contains a framework for performing multi-label classification of Electrocardiogram (ECG) signals using Graph Neural Networks (GNNs). The project is primarily built to process the PTB-XL dataset, a large publicly available ECG dataset containing diverse cardiac pathologies.

Project Structure

• gnn-model-ptbxl.ipynb: The main Jupyter Notebook containing the data-processing pipeline, hyperparameter tuning, model training, and evaluation code.
• requirements.txt: A list of all the Python dependencies required to run the code.

Key Technologies & Libraries

This project uses the following key libraries:

• PyTorch Geometric (torch_geometric): Used to construct and train the Graph Neural Network models for the ECG data.
• WFDB (WaveForm DataBase): For reading and parsing the PTB-XL ECG records.
• NeuroKit2: For advanced physiological signal processing and feature extraction.
• Optuna: Used for hyperparameter optimization to find the best configuration for the GNN.
• Scikit-Learn: For evaluation metrics and data splitting.
• Pandas, NumPy, Matplotlib & Seaborn: For data manipulation, arrays processing, and visualizations.

Getting Started

Prerequisites

You need Python installed on your system. It is highly recommended to use a virtual environment (venv or conda).

Installation

1. Clone this repository:

   git clone https://github.com/VanshGupta18/ECG-Multi-Label-Classification.git
   cd ECG-Multi-Label-Classification

2. Install the required dependencies:

   pip install -r requirements.txt

   Note: For torch_geometric, make sure your PyTorch version and CUDA setup matches the requirements detailed in the PyTorch Geometric installation guide.

3. Download the PTB-XL dataset from PhysioNet and place it in the respective data directories as referenced in the notebook.

Usage

1. Launch Jupyter Notebook or JupyterLab:

   jupyter notebook

2. Open gnn-model-ptbxl.ipynb and run the cells sequentially to build the graph representations of the ECG signals, tune hyperparameters with Optuna, and train/evaluate the GNN multi-label classification model.

Features

• ECG Graph Construction: Converts standard 12-lead ECG signals into graph data structures taking spatial-temporal correlations into account.
• Feature Extraction: Utilizes NeuroKit2 for extracting crucial physiological characteristics.
• GNN Modeling: Uses advanced message passing techniques to predict multiple concurrent cardiac conditions.
• Automated Hyperparameter Tuning: Seamlessly iterates over model configurations using Optuna.

License

This project is licensed under the MIT License.