A simple and fast ANN-based method of studying slow-wave sleep microstructure in freely moving rats

Electroencephalography (EEG) is a common technique for measuring brain activity. Artificial Neuronal Networks (ANNs) can provide valuable insights into the brain dynamics of humans and animals. We built a simple and fast shallow ANN-based solution for sleep recognition in EEGs recorded in freely moving rats. The ANN was constructed using open-source software and truncated to one formula with empirically defined weight coefficients. The optimization of the ANN model's performance (i.e., post-processing) relied on a probability-related approach to sleep microstructure. This approach could be a good way to analyze large datasets. In the current dataset, the slow-wave sleep was recognized with the sensitivity of 0.91 and the specificity of 0.98. The optimal model performance achieved with minimum sleep duration of 80-90 s and sleep interruption of 14-18 s. Our results suggest the following fundamental issues. First, 14-18 s sleep interruptions might be the archetypal micro-arousals in rats. Second, slow-wave sleep in rats might be built up of a set of sleep "building blocks" lasting 80-90 s.