Detection of Cardiac Abnormalities from Multilead ECG using Multiscale Phase Alternation Features

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Electrocardiogram (ECG) morphology changes indicate heart pathology.
• Manual analysis of long-term ECG recordings is challenging for detecting subtle abnormalities.
• Automated ECG analysis is crucial for accurate cardiac abnormality detection.

Purpose of the Study:

• To propose a novel method for automated detection and classification of cardiac abnormalities from multilead ECG signals.
• To utilize multiscale phase alternation (PA) features for enhanced diagnostic accuracy.
• To compare the performance of k-nearest neighbor (KNN) and fuzzy KNN classifiers.

Main Methods:

• Decomposition of multilead ECG signals using dual tree complex wavelet transform (DTCWT) at various scales.
• Computation of phase alternation (PA) values from complex wavelet coefficients as diagnostic features.
• Classification of cardiac abnormalities (BBB, MI, HMD) and healthy controls (HC) using KNN and fuzzy KNN.

Main Results:

• The proposed method achieved high sensitivity for detecting cardiac abnormalities.
• Sensitivity values of 78.12% for BBB, 80.90% for HMD, and 94.31% for MI were reported.
• The fuzzy KNN classifier demonstrated superior performance with the multiscale PA features.

Conclusions:

• The developed multiscale PA features combined with fuzzy KNN offer an effective approach for automated cardiac abnormality detection.
• The method shows promising results for clinical application in diagnosing heart conditions from ECG data.
• Further comparison with state-of-the-art techniques for MI detection highlights the method's potential.