Why the study?
Does data augmentation using GANs improve the accuracy of CNN-based ECG classification in the imbalanced MIT-BIH arrhythmia dataset?
Population
15 different classes of heartbeats from the MIT-BIH arrhythmia dataset (lead 1 only)
Comparison
Data augmentation using Generative Adversarial… vs Training with the original unbalanced dataset…
Design
Other
Key result
Data augmentation using generative adversarial networks (GANs) improved ECG classification by CNNs, achieving overall accuracy >98.0%, precision >90.0%, and sensitivity >97.7%.
Authors
Loading...
May enhance ECG ML research on imbalanced data; leaves open prospective clinical validation before practice use.
Does data augmentation using GANs improve the accuracy of CNN-based ECG classification in the imbalanced MIT-BIH arrhythmia dataset?
Using GANs to balance the MIT-BIH arrhythmia dataset significantly improves the performance of deep learning models for ECG classification, achieving over 98% accuracy.
Shaker et al. (2020) studied Arrhythmias. Data-augmentation using generative adversarial networks (GANs) vs. Original imbalanced dataset and other common data augmentation techniques was evaluated on ECG classification performance (accuracy). Data augmentation using generative adversarial networks (GANs) improved ECG classification by CNNs, achieving overall accuracy >98.0%, precision >90.0%, and sensitivity >97.7%.