Insomnia is prevalent in the general population and is often difficult to be identified reliably. The sleep spindle is a key electroencephalograph (EEG) signal that plays an important role in the preservation of sleep continuity. Previous studies on the relationship between spindle and insomnia mainly focus on the density distribution of spindle waves. In this article, we leverage the large amount of sleep data in the National Sleep Research Resource (NSRR) to develop two sequence models to take into consideration the temporal features of sleep spindles in the whole night sleep recording, and treat the interplay between insomnia and sleep spindle wave as a continuous process. The experimental results on two study cohorts of NSRR show that our method achieved the best performance among all the compared methods, indicating that it is the temporal feature of spindles, rather than stationary features (i.e., frequency, duration, amplitude) that are critical for identifying insomnia patients.
Python3.6
Numpy
Keras
Tensorflow