Detection of cardiac abnormalities from 12-lead ecg using complex wavelet sub-band features

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Cardiac anomaly detection is crucial for diagnosing various heart conditions.
• Electrocardiogram (ECG) signals provide vital information for cardiac health assessment.
• Existing methods for ECG analysis face challenges in accurately identifying subtle abnormalities.

Purpose of the Study:

• To optimize cardiac anomaly detection by identifying the most effective Daubechies wavelet families for ECG signal analysis.
• To differentiate between normal and abnormal cardiac states using machine learning algorithms on ECG data.
• To enhance the precision of cardiac irregularity identification using advanced signal processing techniques.

Main Methods:

• A novel method combining Discrete Wavelet Transform (DWT) with Principal Component Analysis (PCA) for feature extraction and dimension reduction from ECG signals.
• Utilizing various Daubechies wavelet families (db2-db6) for signal decomposition.
• Employing machine learning classifiers including Multilayer Perceptron (MLP) neural network, Ensemble Subspace K-Nearest Neighbour (KNN), and Ensemble Bagged Trees for classification.

Main Results:

• The Daubechies wavelet family db5 demonstrated superior performance in representing ECG signals.
• The proposed method effectively distinguished between normal and abnormal ECG signals in the Physionet Apnea ECG database.
• The MLP neural network achieved 100% accuracy for healthy signals and 97.8% for abnormal signals, with an overall accuracy of 98.6%.

Conclusions:

• The selection of an appropriate Daubechies wavelet family is critical for accurate ECG analysis.
• Machine learning techniques, particularly neural networks, combined with wavelet transforms, significantly improve cardiac abnormality detection.
• The developed method shows high potential for clinical application in diagnosing cardiac irregularities.