Thalamo-cortical coupling during N2 sleep spindles: Combining MEG and machine learning

Sleep Spindles are scalp EEG waveforms that are hallmarks of the N2 sleep stage and are believed to be generated through thalamo-cortical loops. In this study we set out to study the interaction between cortical and subcortical regions involved in spindle activity using non-invasive measurements. To achieve this goal, we used, (a) source reconstruction of MEG data, (b) functional connectivity (inter-regional coherence) in order to model the interaction between cortical and subcortical structures in the typical spindle frequency band (sigma, 11-16 Hz) and a control band (theta, 4-7 Hz) both during spindles and baseline, (c) data driven analyses using supervised machine learning techniques in order to identify the interaction variables that are most relevant for spindle activity. Our results shows that oscillatory interaction metrics between thalamus and central or frontal cortex provide significant discrimination (spindle vs non-spindle windows) reaching decoding accuracies of 81,81% in the sigma band (hippocampus-cortical coupling also led to successful decoding reaching up to 77,72% in the sigma band). These findings suggest that detecting activity from the thalamus and hippocampus is possible using MEG data, and in particular that MEG allows us to monitor the oscillatory coupling properties thought to mediate spindles during sleep. These results will allow us to better the involvement of deep brain structures in the functional role of sleep spindles