A Spectral-Based Approach for BCG Signal Content Classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Physiological Monitoring

Background:

• Ballistocardiogram (BCG) signals are crucial for non-invasive vital sign measurement.
• Human body activities frequently interfere with BCG signal quality, complicating heart rate (HR) and respiratory rate (RR) extraction.
• Accurate segmentation of BCG signals based on activity is needed to improve vital sign measurement reliability.

Purpose of the Study:

• To develop binary flags identifying BCG signal frames suitable for HR and RR measurement.
• To achieve refined BCG signal segmentation by classifying various human body activities.
• To enhance the accuracy and reliability of vital sign extraction from BCG signals.

Main Methods:

• Utilized unsupervised Gaussian Mixture Model (GMM) classification for BCG signal exploration and hyper-parameter definition.
• Employed supervised K-Nearest Neighbors (KNN) classification for frame-level and temporal series analysis of spectral features (SFM, SC).
• Classified human body activities (e.g., coughing, breath holding, movement) from BCG signals using two-level supervised classification.

Main Results:

• Successfully generated binary flags to indicate useful BCG signal frames for vital sign measurement.
• Achieved high accuracy (94.6%) in segmenting BCG signals according to distinct human body activities.
• Demonstrated the effectiveness of spectral features (SFM, SC) and temporal series analysis for activity classification.

Conclusions:

• The proposed framework offers a novel and powerful method for BCG signal segmentation based on human body activities.
• This approach significantly improves the potential for reliable cardiac and respiratory rate measurement from BCG signals.
• The developed technique addresses a key challenge in BCG signal processing, paving the way for more robust vital sign monitoring.