A novel adaptive time-window method for detecting slow wave–spindle coupling: Comparison of temporal co-occurrence and phase-amplitude coupling approaches

Abstract

Background

Slow wave–spindle (SW-SP) coupling is critical to the role of sleep in cognition. However, the reliability and validity of available detection methods remain lacking, employ varying approaches, and have yet to be directly compared and rigorously validated. This study aimed to: (1) compare phase amplitude coupling (PAC)-based methods (PACTools and YASA) with the “temporal co-occurrence”, Fixed Time Window approach for detecting SW-SP coupling, and, (2) introduce a refined adaptive SW-SP coupling method to improve precision and accuracy, by precisely aligning spindle detection with slow wave half-wave durations.

New method

A novel SW-SP coupling detection method was developed, incorporating dynamic time windows based on slow wave half-wave durations. This method adjusts to the morphology of individual slow waves offering precise alignment of spindle events with slow wave peaks and troughs.

Results

PAC-based methods showed high sensitivity but low specificity, resulting in excessive false positives. Across PAC indices (MVLMI, KLMI, PLV, GLMMI), the average F1-score was ∼0.45 ± 0.01. YASA, which detects coupling events based on sigma-band oscillations, exhibited moderate accuracy (0.46 ± 0.007), with an F1-score of 0.446 ± 0.011, reflecting its tendency to over-detect events due to reliance on sigma power fluctuations rather than discrete spindles. The Fixed Time Window method demonstrated higher specificity, identifying 533 ± 28 coupled spindles per participant, but relied on static temporal boundaries, leading to an average lag of 1.04 ± 0.01 s relative to slow wave peaks. The Adaptive Half-Wave method improved upon this by dynamically adjusting detection windows to slow wave morphology, reducing the average lag to 0.15 ± 0.007 s while maintaining high specificity (accuracy = 0.83 ± 0.01, precision = 0.96 ± 0.01).

Comparison with existing methods

The adaptive method provides a significant improvement in temporal precision and specificity by dynamically aligning detection windows with slow wave morphology. Fully integrated into the Counting Sheep PSG EEG toolbox, it streamlines workflows for spindle detection, slow wave characterization, and coupling analysis within an easy to use, EEGLAB-compatible environment.

Conclusions

As SW-SP coupling gains recognition as an important measure of sleep’s role in cognition, the need for a standardized detection method has become clear. The adaptive method provides a robust, open-source solution, addressing the need for standardized SW-SP coupling detection.