ECG beat classification by a novel hybrid neural network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiogram (ECG) signal analysis is crucial for diagnosing cardiac conditions.
• Accurate classification of ECG beats is essential for automated cardiac monitoring.
• Existing neural network models face challenges in efficiency and accuracy.

Purpose of the Study:

• To propose a novel hybrid neural network for enhanced ECG beat classification.
• To compare the performance of the hybrid network with traditional Multi-Layer Perceptron (MLP) and Restricted Coulomb Energy (RCE) models.
• To investigate the impact of feature extraction methods (Fourier and wavelet analyses) on classification accuracy.

Main Methods:

• Development of a hybrid neural network integrating Fourier and wavelet feature extraction.
• Utilizing dynamic programming for ECG feature determination based on divergence values.
• Training the hybrid network using genetic algorithms (GAs) to optimize performance and reduce node count.
• Comparative analysis of classification performance, training time, and network complexity against MLP and RCE models.

Main Results:

• The novel hybrid neural network achieved a 96% classification success rate for ten types of ECG beats.
• The hybrid structure demonstrated improved classification performance and reduced the number of nodes compared to MLP and RCE.
• Genetic algorithms effectively enhanced the hybrid network's classification accuracy and efficiency.
• Both Fourier and wavelet analyses proved effective for ECG beat feature extraction in an eight-dimensional space.

Conclusions:

• The proposed hybrid neural network offers a superior approach for ECG beat classification.
• Genetic algorithm optimization significantly enhances the efficiency and accuracy of the hybrid model.
• This method holds promise for real-time ECG monitoring and cardiac diagnostics.