ScaHybNet: a scalogram-based hybrid ensemble network for ECG arrhythmia classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Cardiovascular diseases are a leading global cause of mortality.
• Accurate and timely detection of cardiac arrhythmias is crucial for preventing sudden cardiac death.
• Existing methods may struggle with noisy and imbalanced electrocardiogram (ECG) data.

Purpose of the Study:

• To propose ScaHybNet, a novel deep learning ensemble model for multi-class arrhythmia classification.
• To enhance the accuracy and robustness of arrhythmia detection using ECG signals.
• To address challenges posed by class imbalance and noise in ECG datasets.

Main Methods:

• ECG signals were transformed into 224x224 RGB-scalogram images using Continuous Wavelet Transform (CWT) with a Morlet wavelet.
• A hybrid deep learning architecture combining a Convolutional Neural Network (CNN) with residual blocks, a Bidirectional Long Short-Term Memory (BiLSTM) layer, and a Transformer encoder was developed.
• Stratified balancing and inverse-frequency class weighting were employed to mitigate extreme class imbalance.
• Fivefold cross-validation was used to assess model robustness.

Main Results:

• ScaHybNet achieved an ensemble training accuracy of 99.81%.
• The mean accuracy across fivefold cross-validation was 90.42% ± 1.26%.
• On the unseen test set, the model demonstrated an ensemble test accuracy of 94.73%, with a precision of 76.51%, recall of 82.93%, and F1-score of 77.40%.

Conclusions:

• ScaHybNet effectively classifies cardiac arrhythmias, showing superior or comparable performance to state-of-the-art methods, particularly with noisy and imbalanced ECG data.
• The model's hybrid architecture, incorporating CNN, BiLSTM, and Transformer components, proves effective in learning spatial, temporal, and long-term dependencies.
• ScaHybNet holds significant potential as a patient-centric tool to benefit the medical field in cardiovascular disease management.