Patching with Sequential Updating for High-Fidelity Bayesian Spectral Estimation of Physiological Time Series

Area of Science:

• Physiological signal processing
• Biomedical data analysis
• Time series analysis

Background:

• Physiological time series offer insights into system behavior.
• Analyzing imperfect physiological data with gaps is challenging.
• Existing spectral estimation methods struggle with fragmented signals.

Purpose of the Study:

• To introduce a novel method, Patching with Sequential Updating for Bayesian Nonparametric Spectral Estimation (PBNSE).
• To enhance spectral estimation and interpretation of fragmented physiological time series.
• To address challenges like irregular sampling, incomplete signals, and varying noise.

Main Methods:

• PBNSE models data segments as patch-specific Gaussian processes (GPs).
• It employs patch-dependence with a joint GP and shared kernel for cross-patch dependencies.
• Sequential parameter shifting transfers knowledge between patches, enabling unified power spectral density (PSD) estimation.

Main Results:

• PBNSE demonstrated significant improvements in spectral accuracy and robustness.
• It outperformed state-of-the-art methods including BNSE, multitaper, and Lomb-Scargle.
• The method effectively handles irregular sampling, incomplete signals, and noise.

Conclusions:

• PBNSE offers a robust solution for analyzing imperfect physiological time series.
• The method enhances spectral estimation accuracy and interpretation robustness.
• Widespread adoption can advance physiological signal research and real-world applications.