AI based ECG data recovery and cardiovascular diseases classification (CEDRC-network)

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of mortality, necessitating accurate and timely diagnosis.
• Electrocardiograms (ECG) are crucial for detecting heart conditions but can be compromised by noise and missing data.
• Existing diagnostic methods struggle with non-ideal ECG signals, impacting patient outcomes.

Purpose of the Study:

• To develop an advanced AI system, the Cardiovascular ECG Data Recovery and Classification Network (CEDRC-Net), for robust ECG signal processing.
• To enhance the accuracy of CVD diagnosis by effectively handling noisy and incomplete ECG data.
• To improve patient recovery and longevity through more reliable and timely cardiac condition identification.

Main Methods:

• The CEDRC-Net integrates a multistage machine learning and deep learning model, incorporating a Transformer-based Convolutional Denoising Autoencoder (TCDAE).
• Noise reduction is achieved using TCDAE, followed by ECG signal reconstruction and forecasting via Variational Autoencoder (VAE) or Temporal Fusion Transformer (TFT).
• Classification of heart diseases (atrial fibrillation, sinus bradycardia, tachycardia) is performed using machine learning algorithms on processed ECG data from 2426 patients.

Main Results:

• The Temporal Fusion Transformer (TFT) demonstrated superior ECG signal reconstruction accuracy (98%) compared to Variational Autoencoder (VAE) (96%).
• ECG signals enhanced by TFT resulted in higher classification accuracy (98.4% with SVM and XGBoost) and F1-scores.
• TFT achieved significantly lower reconstruction errors (MAE=0.015, MSE=0.00045, RMSE=0.0132) than VAE (MAE=0.075, MSE=0.011, RMSE=0.107).

Conclusions:

• The CEDRC-Net, particularly with TFT, offers strong denoising and reconstruction capabilities, significantly improving ECG diagnostic accuracy.
• The system provides reliable ECG measurements under noisy conditions, facilitating early and accurate diagnosis.
• This advancement supports better clinical decisions, reduces patient load on cardiologists, and contributes to improved patient longevity.