Ventricular ectopic beat detection using a wavelet transform and a convolutional neural network.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Ventricular ectopic beats (VEBs) disrupt normal cardiac function and heart rate variability analysis.
• VEBs can be misidentified as artifacts due to similar morphology and timing.
• Accurate VEB detection is crucial for identifying potentially life-threatening cardiac conditions.

Purpose of the Study:

• To develop and evaluate a method for automated classification of ventricular ectopic beats (VEBs).
• To utilize wavelet transform and deep learning for enhanced VEB identification.
• To assess the method's performance on independent cardiac arrhythmia databases.

Main Methods:

• Electrocardiogram (ECG) segments were transformed into 2D time-frequency images using three wavelet types (Morlet, Paul, Gaussian derivative).
• A convolutional neural network (CNN) was employed to classify these images, optimizing filters for accuracy.
• The model was validated using ten-fold cross-validation on the MIT-BIH arrhythmia database and tested on the AHA database.

Main Results:

• The proposed algorithm, particularly with the Paul wavelet, achieved an 84.94% F1 score and 97.96% accuracy on the MIT-BIH dataset.
• Independent testing on the AHA database yielded an 84.96% F1 score and 97.36% accuracy.
• The results demonstrate robust performance and effective discrimination of VEBs from other cardiac beats and artifacts.

Conclusions:

• The combination of wavelet transform and CNN provides an effective automated method for VEB classification.
• The developed network exhibits strong generalization capabilities, performing well across different datasets.
• This approach offers a significant advancement in the accurate and reliable detection of ventricular ectopic beats.