Key Points
- Develop and evaluate a trainable neural network ensemble system to improve electrocardiogram beat classification and identify premature ventricular contractions.
- Applied stationary wavelet transform to reduce background noise in electrocardiogram recordings sourced from the MIT-BIH arrhythmia database.
- Extracted 10 morphological wave features and one timing interval feature from each processed beat.
- Trained multiple multilayer perceptron neural network architectures and integrated them via stacked generalization.
- Stacked generalization ensemble achieved the highest recognition rate among evaluated models at approximately 95%.
- The integrated pipeline effectively differentiated premature ventricular contractions from normal beats and other cardiac abnormalities.
Structured PICO
PPopulationECG samples attributing to different ECG beat types extracted from the MIT-BIH arrhythmia database
IInterventionTrainable neural network ensemble approach (Stacked Generalization) with stationary wavelet transform (SWT) denoising and extraction of 10 morphological and 1 timing interval features
CComparatorOther neural network combination methods and multilayer perceptron (MLP) topologies
OOutcomeRecognition rate for classification of ECG beats (detection of premature ventricular contraction from normal beats and other heart diseases)
A trainable neural network ensemble using Stacked Generalization achieves a 95% recognition rate for ECG beat classification, demonstrating potential for automated arrhythmia diagnosis systems.