Subband higher-order statistics and cross-correlation for heartbeat type recognition based on two-lead...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional electrocardiogram (ECG) beat classification relies on single-lead signals.
• This limits recognition rates and noise tolerance in diagnostic systems.

Purpose of the Study:

• To investigate the impact of incorporating a second ECG lead on classification accuracy.
• To evaluate the enhanced noise-tolerance of a two-lead ECG system.

Main Methods:

• Utilized discrete wavelet transformation for ECG signal decomposition.
• Applied higher-order statistics and inter-lead cross-correlation for feature extraction.
• Employed a feed-forward back-propagation neural network (FFBNN) as the classifier.

Main Results:

• The two-lead system increased recognition rates from 97.74% to 98.25% compared to a single-lead system.
• Maintained over 97.8% accuracy even at a signal-to-noise ratio (SNR) of 10 dB.
• Demonstrated superior noise-resistibility in heartbeat discrimination.

Conclusions:

• A second ECG lead significantly enhances the performance of automated heartbeat classification systems.
• The two-lead approach offers improved accuracy and robustness against noise, crucial for clinical applications.