Unsupervised Sleep State Detection in Mice Using a Physiologically-Based Clustering Algorithm

In order to perform high throughput studies of sleep in animals, it is desirable to automate the detection of different sleep states. Inspiration is drawn from a variety of other sleep staging algorithms to create an unsupervised model for sleep staging in mice. First, a state space is created using physiologically relevant features including oscillatory activity in the dorsal hippocampus and muscle activity in the trapezius muscle. A density-based visualization of the state space reveals three distinct clusters, which demonstrate the expected physiological properties of wakefulness, REM sleep, and non-REM sleep. Finally, a GMM-based clustering method that requires minimal user interface is applied to the state space to extract sleep state information. Although there is much work to be done, the clusters match human intuition and are a promising step towards a fully automated sleep staging algorithm.

To read more, please click on the link below

GWDDECE_Hefter