A user’s introduction to an algorithmic method to identify space-time profiles of sleep slow oscillations: dataset constraints, case-use examples, and open-source code.

Abstract Studies of sleep slow oscillations (SOs, 0.5-1.5 Hz) have emphasized their importance for cognition and health, and their variable spatial organization. We have introduced a data-driven method to analyze SOs as events that differentiate in their space-time co-emergence on the electrode manifold. This approach has identified properties of SO organization that are relevant to function, and that can change in clinical populations. In this work, we share a software and user manual that will allow the sleep research community to leverage our method directly in their own datasets. The work formalizes which dataset properties are necessary to deploy our method in terms of number of participants (N) and count of electrodes (E), and share parameterization strategies. We applied our algorithm to two datasets of nighttime sleep in healthy adults: Set1 (N=22, E=58) and Set 2 (N=34, E=24). Roles of E and N values were tested by down-sampling electrodes to 24 and 8 channels, reflecting standard caps, and by randomly selecting subsets of participants. Early vs complete nighttime sleep was evaluated by truncating sets to 90 minutes after the first detected SO. Clustering outputs from tests were compared to original dataset outputs. Successful identification of SO profiles was evaluated with an index of similarity to ideal centroid masks. We found that identification of SO profiles required at least 22 participants and at least a 24 head-electrode montage, whereas 8 head-electrodes configurations, typical of clinically acquired sleep, were not sufficient. Furthermore, early nighttime sleep was sufficient for successful identification of SO profiles.