0353 Automated Synchronization of Wrist-Worn Sleep Data with Polysomnography Using Heart Rate Normalized Cross-Correlation

Abstract Introduction Accurate validation of sleep staging algorithms from wrist-worn wearable sensors against polysomnography (PSG) requires precise temporal alignment between the two datasets. Traditional manual synchronization methods, such as event marker presses or device taps, are prone to errors, burdensome for participants and study sites, and labor-intensive for post-processing. The advent of photoplethysmography (PPG) sensors in modern wrist-worn devices enables reliable heart rate estimation, providing a robust physiological signal for automated alignment. Herein, we propose a method to synchronize wrist-worn wearable data with PSG using heart rate signals from both systems. Methods We developed an automated synchronization approach leveraging heart rate (HR) signals from a wrist-worn wearable and a 4-lead electrocardiogram (ECG) during PSG. HR data from both sources were aggregated into 1-second medians. Normalized cross-correlation analysis was performed by shifting the wearable HR signal relative to the ECG signal in 1-second increments over an adjustable ±60-second window, yielding 121 normalized correlation coefficients. The lag maximizing the normalized cross-correlation served as the synchronization offset. To mitigate artifacts from device processing errors or R-peak detection inaccuracies, only physiologically plausible HR values (30–200 bpm) were included in the correlation computation. Results In simulated datasets with induced artificial lags ranging from -30 to +30 seconds, the method recovered offsets with near-perfect accuracy. Visualizations of pre- and post-synchronization HR traces demonstrated tight alignment, with normalized cross-correlation plots exhibiting sharp peaks at the true offset, confirming robust detection even amid noise. Application to real validation datasets (n=30 nights) achieved clear peaks in the normalized cross-correlation output and reduced manual synchronization time by over 80%. Conclusion Normalized cross-correlation of HR signals provides a reliable, automated method for synchronizing wrist-worn wearable data with PSG, enhancing the efficiency and accuracy of sleep staging validation studies. This approach minimizes human error and scales well for large cohorts, paving the way for broader adoption of wearables in clinical sleep research. Support (if any) Ametris supported this work