Assessment of blind source separation techniques for video-based cardiac pulse extraction

Area of Science:

• Biomedical Engineering
• Signal Processing
• Physiological Monitoring

Background:

• Contactless vital sign acquisition using camera-based photoplethysmogram (cbPPG) relies heavily on blind source separation (BSS) for cardiac pulse extraction.
• Optimal BSS application and its general benefits for cbPPG analysis remain debated.
• Understanding BSS performance under varying input conditions is crucial for reliable contactless vital sign monitoring.

Purpose of the Study:

• To investigate the performance of BSS techniques in enhancing cardiac pulse signals from cbPPGs.
• To evaluate how different input data characteristics affect BSS performance.
• To compare the efficacy of Principal Component Analysis (PCA) and Independent Component Analysis (ICA) for cbPPG enhancement.

Main Methods:

• Automated spatial preselection of regions of interest to control BSS input conditions.
• Processing cbPPG data from 18 cardiovascular patients with varying characteristics (wavelength, dominant frequency, signal quality).
• Application of standard BSS techniques: Principal Component Analysis (PCA) and Independent Component Analysis (ICA).

Main Results:

• Preselection of cbPPG regions of interest significantly improved spectral signal-to-noise ratio (SNR) compared to standard methods.
• Both PCA and ICA demonstrated superior performance with monochrome (green wavelength) inputs versus multi-wavelength inputs.
• PCA outperformed ICA for more homogeneous input signals.
• High input SNR with standard contrast ICA application potentially decreased the cardiac pulse SNR.

Conclusions:

• Automated spatial preselection is a beneficial preprocessing step for cbPPG analysis.
• Monochrome green light acquisition is advantageous for BSS-based cardiac pulse extraction.
• PCA is generally more effective than ICA for homogeneous cbPPG signals, especially under high SNR conditions.