Automatic Detection of Atrial Fibrillation Based on Continuous Wavelet Transform and 2D Convolutional Neural Networks

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Atrial fibrillation (AF) is a prevalent cardiac arrhythmia linked to significant morbidity and mortality.
• Early and automatic detection of AF remains a challenge, hindering effective treatment strategies.
• Current methods often focus on isolated atrial or ventricular activity, limiting detection capabilities.

Purpose of the Study:

• To develop a novel, accurate, and automated method for detecting atrial fibrillation (AF) episodes.
• To leverage time-frequency features of electrocardiogram (ECG) signals for improved AF detection.
• To establish a new benchmark for AF detection performance using advanced machine learning.

Main Methods:

• A new method combining continuous wavelet transform (CWT) and 2D convolutional neural networks (CNNs) was developed.
• The algorithm analyzes the time-frequency characteristics of ECG signals.
• The MIT-BIH Atrial Fibrillation Database was utilized for training and validation.

Main Results:

• The proposed algorithm achieved high performance metrics: 99.41% sensitivity, 98.91% specificity, 99.39% positive predictive value, and 99.23% overall accuracy.
• The method demonstrated efficacy compared to existing algorithms on the same dataset.
• The algorithm can detect AF episodes using as few as five ECG beats.

Conclusions:

• The developed CWT and 2D CNN-based method offers a highly accurate and efficient approach for AF detection.
• Analyzing time-frequency ECG features provides an advantage over traditional methods.
• The algorithm's ability to detect AF from minimal data suggests significant potential for practical clinical applications.