Skip to content

Automatic ECG classification using discrete wavelet transform and one-dimensional convolutional neural network

License

Notifications You must be signed in to change notification settings

arminshoughi/ecg-classification-cnn

Repository files navigation

ECG Signal Classification with Deep Learning

Overview

Welcome to the repository for the implementation of our paper on accurate Electrocardiogram (ECG) signal classification using deep learning. ECG signals play a vital role in providing crucial cardiovascular information for medical practitioners. Manual analysis of these signals is intricate and time-consuming, requiring specific skills. The challenges posed by noise, signal rigidity, and irregular heartbeats make it essential to employ advanced techniques for accurate classification.

Approach

Our approach leverages a Convolutional Neural Network (CNN), discrete wavelet transformation with db2 mother wavelet, and the Synthetic Minority Over-sampling Technique (SMOTE). We applied this methodology to the MIT-BIH dataset, adhering to the Association for the Advancement of Medical Instrumentation (AAMI) standards. The aim is to enhance the accuracy of ECG signal classifications, particularly for cardiovascular diseases (CVDs), a leading global cause of mortality.

Results

After training for 50 epochs, with each epoch taking 39 seconds, our approach achieved remarkable accuracy:

  • Category F: 99.71%
  • Category N: 98.69%
  • Category S: 99.45%
  • Category V: 99.33%
  • Category Q: 99.82%

These results demonstrate the effectiveness of our model in accurately classifying ECG signals, making it a potential clinical auxiliary diagnostic tool.

Usage

To use the code, clone the repository:

git clone https://gitlab.com/arminshoughi/ecg-classification-cnn
cd ecg-classification-cnn
pip install -r requirements.txt
python EcgClassification.py

Follow the installation steps and guidelines in the codebase to replicate our experiments and apply the model to your own datasets.

Evaluation

For a detailed understanding of our methodology and evaluation results, please refer to our published paper. The complete article is available here.

Contact

For any inquiries or collaboration opportunities, feel free to contact the project maintainer, Armin Shoughi, via the repository issues.

Thank you for your interest in our work!

Releases

No releases published

Packages

No packages published