Arrhythmias Classification Using Short-Time Fourier Transform and GAN Based Data Augmentation.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Accurate classification of cardiac arrhythmias using artificial neural networks (ANNs) is hindered by insufficient training data.
• Limited availability of diverse heart rhythm samples impedes the development of robust arrhythmia detection models.

Purpose of the Study:

• To develop a novel data augmentation method to address the scarcity of training samples for arrhythmia classification.
• To enhance the accuracy of artificial neural network models for detecting various heart rhythm abnormalities.

Main Methods:

• A data augmentation technique combining short-time Fourier transform (STFT) and generative adversarial networks (GANs) was proposed.
• Electrocardiogram (ECG) signals were transformed into coefficient matrices using STFT.
• GANs were trained on these matrices to generate synthetic data for augmenting training datasets for convolutional neural networks (CNNs).

Main Results:

• The proposed STFT-GAN data augmentation strategy significantly improved the performance of all tested CNN-based classification models.
• Augmented datasets led to enhanced accuracy in classifying different heart rhythms.
• The method demonstrated effectiveness in addressing the challenge of limited data for biomedical signal analysis.

Conclusions:

• The novel STFT-GAN data augmentation method effectively increases the number of evenly distributed training samples for arrhythmia classification.
• This approach offers a valuable solution for improving the accuracy of ANNs in detecting multiple arrhythmias, particularly when adequate real-world data is scarce.
• The proposed technique shows promise for broader applications in augmenting and classifying biomedical signals.