Intelligent arrhythmia detection and classification using ICA.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Cardiac arrhythmias pose a significant health risk, necessitating accurate and efficient detection methods.
• Current detection techniques often face challenges in feature extraction and classification accuracy.
• Automated analysis of electrocardiogram (ECG) signals is crucial for early diagnosis.

Purpose of the Study:

• To propose a novel approach for cardiac arrhythmia detection using advanced signal processing and machine learning.
• To evaluate the effectiveness of feature extraction techniques like independent component analysis (ICA) and wavelet transform.
• To compare the performance of different machine learning classifiers for arrhythmia classification.

Main Methods:

• Feature extraction using independent component analysis (ICA) and wavelet transform.
• Classification of 274 ECG signals from the MIT-BIH database using Multilayer Perceptron (MLP), Radial Basis Function (RBF) neural networks, and K-nearest neighbor (KNN).
• Optimization of the learning system using the Levenberg-Marquardt (LM) backpropagation algorithm for multilayer neural networks.

Main Results:

• Multilayer neural networks (MLP) with the Levenberg-Marquardt (LM) algorithm demonstrated optimal performance.
• The proposed method achieved a high accuracy of 98.5% in classifying cardiac arrhythmias.
• The results show a significant improvement compared to similar existing works in the field.

Conclusions:

• The combined approach of ICA and wavelet transform for feature extraction is highly effective for cardiac arrhythmia detection.
• MLP neural networks, optimized with the LM algorithm, provide a robust and accurate classification system.
• This novel method offers a promising advancement for automated cardiac arrhythmia diagnosis.