Application of higher order spectra for accurate delineation of atrial arrhythmia

Area of Science:

• Cardiology and Biomedical Signal Processing
• Non-linear dynamics and statistical analysis of physiological signals

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing atrial tachyarrhythmias, but visual interpretation often leads to misdiagnoses due to signal complexity.
• Traditional time and frequency domain analyses of ECG signals may not capture all relevant diagnostic information, particularly for non-linear and non-stationary variations.
• Accurate differentiation between normal sinus rhythm, atrial fibrillation, and atrial flutter is clinically significant for appropriate patient management.

Purpose of the Study:

• To develop and evaluate a novel methodology for ECG-based pattern analysis to detect and differentiate between normal sinus rhythm, atrial flutter, and atrial fibrillation.
• To investigate the utility of non-linear features derived from higher-order spectra (HOS) for classifying these cardiac arrhythmias.
• To assess the performance of automated classifiers (KNN, CART, NN) following data reduction using Independent Component Analysis (ICA) of bispectrum features.

Main Methods:

• Extraction of non-linear features from higher-order spectra (HOS), specifically bispectrum, from ECG signals.
• Application of Independent Component Analysis (ICA) for dimensionality reduction of the extracted bispectrum features.
• Classification of ECG beats using K-nearest-neighbor (KNN), Classification and Regression Tree (CART), and Neural Network (NN) algorithms.

Main Results:

• The proposed method achieved an average accuracy of 97.65% in classifying normal sinus rhythm, atrial fibrillation, and atrial flutter beats.
• High sensitivity (98.75%) and specificity (99.53%) were obtained using ten-fold cross-validation, demonstrating robust classification performance.
• The study confirmed that higher-order spectra statistics are effective for classifying atrial tachyarrhythmias with high accuracy.

Conclusions:

• Non-linear analysis of ECG signals using higher-order spectra provides a powerful tool for accurate detection of atrial tachyarrhythmias.
• The combination of HOS features, ICA, and machine learning classifiers offers a promising automated approach for clinical diagnosis.
• Further validation is recommended for potential clinical implementation of this advanced ECG analysis technique.