Wavelet Scattering Transform for ECG Beat Classification

Area of Science:

• Cardiology
• Signal Processing
• Machine Learning

Background:

• Electrocardiograms (ECG) contain vital information for diagnosing cardiovascular diseases like arrhythmia.
• Analyzing complex and nonlinear ECG signals visually is challenging.
• Wavelet scattering transform offers stable signal representations.

Purpose of the Study:

• To develop an automated method for classifying four categories of arrhythmia ECG heartbeats: nonectopic (N), supraventricular ectopic (S), ventricular ectopic (V), and fusion (F) beats.
• To evaluate the effectiveness of wavelet scattering transform combined with machine learning classifiers for ECG analysis.

Main Methods:

• Utilized wavelet scattering transform to extract 8 time windows from ECG heartbeats.
• Applied dimensionality reduction techniques: Principal Component Analysis (PCA) and time window selection.
• Classified features using Neural Network (NN), Probabilistic Neural Network (PNN), and K-Nearest Neighbour (KNN) classifiers.

Main Results:

• The 4th time window combined with KNN (k=4) yielded optimal classification performance.
• Achieved an average accuracy of 99.3%, positive predictive value of 99.6%, sensitivity of 99.5%, and specificity of 98.8% via tenfold cross-validation.
• Demonstrated the model's capability for highly accurate arrhythmia classification.

Conclusions:

• The proposed wavelet scattering transform-based model accurately classifies arrhythmia ECG heartbeats.
• This automated approach can assist physicians in interpreting ECG signals, improving diagnostic efficiency.