Standardized synchronization and validation pipeline for physiological biomarkers across multiple devices

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Physiological Signal Processing

Background:

• Continuous physiological monitoring via wearables is crucial for real-world health studies.
• A lack of standardized methods hinders data synchronization across different wearable devices.
• This limits the integration and comparability of data in longitudinal and multi-site research.

Purpose of the Study:

• To present a generalizable, user-friendly pipeline for synchronizing physiological signals from any two wearable devices.
• To validate the pipeline by comparing the Empatica E4 and EmbracePlus devices using concurrent recordings.
• To assess signal-level agreement and identify device-specific variability for key physiological signals.

Main Methods:

• Developed a non-coding pipeline involving resampling, dynamic time warping, wavelet-based amplitude correction, and standardization.
• Applied the pipeline to 48-hour concurrent recordings from 31 participants comparing Empatica E4 and EmbracePlus.
• Analyzed agreement using correlation coefficients, Bland-Altman plots, RMSE, KL divergence, spectral coherence, mutual information, and feature-level correlations via NeuroKit2 and FLIRT.

Main Results:

• Near-perfect agreement (CCC ≈ 1.0) observed for blood volume pulse (BVP) and high agreement for phasic electrodermal activity (EDA) features (CCC 0.85-0.99).
• Tonic EDA, temperature, and accelerometry showed systematic amplitude biases and axis-dependent variability, with lower agreement on certain axes.
• Relative signal dynamics were preserved across devices, despite absolute amplitude differences.

Conclusions:

• The developed pipeline effectively synchronizes wearable physiological data, enabling integration of BVP, EDA, and temperature signals between Empatica E4 and EmbracePlus.
• Findings support the interchangeability of BVP, EDA, and TEMP data with appropriate preprocessing.
• Calibration is recommended for accelerometry data due to observed device-specific variability and amplitude biases.