Source separation from single-channel abdominal phonocardiographic signals based on independent component analysis

Area of Science:

• Biomedical Engineering
• Signal Processing
• Maternal-Fetal Medicine

Background:

• Continuous fetal heart rate (FHR) monitoring is crucial for diagnosing fetal heart abnormalities.
• Fetal phonocardiogram (fPCG) signals, acquired noninvasively, are often contaminated by maternal heart sounds and environmental noise, resulting in low signal-to-noise ratio (SNR).
• fPCG signals are nonstationary, with characteristics varying based on gestational age, fetal position, maternal factors, and recording system bandwidth.

Purpose of the Study:

• To develop and evaluate a source separation technique for extracting FHR from single-channel fPCG signals.
• To address the challenges of low SNR and non-stationarity in fPCG recordings.
• To improve the accuracy of FHR measurement for better fetal heart function evaluation.

Main Methods:

• Proposed a source separation process for single-channel fPCG data.
• Implemented and compared three combined Independent Component Analysis (ICA) methods: SSA-ICA, Wavelet-ICA, and EEMD-ICA.
• Decomposed single-channel fPCG signals into multivariate data prior to ICA application.

Main Results:

• The SSA-ICA method demonstrated effectiveness in recovering sources from single-channel fPCG signals with varying SNRs.
• Performance validation was conducted using both simulated and real fPCG signals.
• Quantitative analysis using power spectral density (PSD) peak and cross-correlation values supported the superiority of SSA-ICA.

Conclusions:

• The SSA-ICA method is more successful in extracting independent sources from fPCG signals compared to Wavelet-ICA and EEMD-ICA.
• This technique offers a promising approach for clearer FHR extraction from noisy, nonstationary fPCG data.
• The findings contribute to improved non-invasive fetal cardiac monitoring and diagnosis.