A Feasible Feature Extraction Method for Atrial Fibrillation Detection From BCG

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Atrial fibrillation (AF) is a prevalent arrhythmia causing significant health issues.
• Accurate AF detection is crucial for patient quality of life and disease management.
• Non-contact physiological signal monitoring offers a promising avenue for continuous cardiac assessment.

Purpose of the Study:

• To develop and validate a feature extraction method for AF detection using ballistocardiogram (BCG) signals.
• To assess the efficacy of machine learning algorithms in classifying AF and sinus rhythm (SR) based on BCG features.
• To explore the potential of BCG-based methods for long-term home cardiac monitoring and AF screening.

Main Methods:

• Collected overnight BCG signals from 37 subjects (37 with AF, 37 with SR).
• Processed 1-minute BCG segments, transforming them into BCG energy signals.
• Extracted 16 features (mean, variance, skewness, kurtosis) from four derived data sequences.
• Classified AF and SR using five machine learning algorithms, with a focus on Support Vector Machine (SVM).

Main Results:

• The Support Vector Machine (SVM) classifier demonstrated superior performance.
• SVM achieved a sensitivity of 0.968, precision of 0.928, and accuracy of 0.945.
• The proposed feature extraction method effectively distinguished between AF and SR using BCG signals.

Conclusions:

• The developed BCG feature extraction method is effective for AF and SR classification.
• This approach holds potential for developing non-contact systems for long-term home cardiac monitoring.
• The findings lay the groundwork for future AF screening tools utilizing BCG technology.