Sleep Stage Probabilities Derived from Neurological or Cardiorespiratory Signals by Means of Artificial Intelligence

The gold standard for scoring sleep according to the rules defined by the American Academy of Sleep Medicine (AASM) relies on human expert scoring based on neurological signals. However, there is a current move from visual scoring toward automated scoring of sleep stages, since the manual scoring process is time-consuming, error prone, and can be performed only by well-trained and experienced human scorers with nevertheless limited interrater reliability. Recent years have seen the maturing of artificial intelligence (AI) algorithms that take on the scoring task, offering consistent and reliable scoring and additional features such as estimated sleep stage probabilities for each epoch (hypnodensity graph). Of particular interest, given the increasing trend from attended in-lab full night polysomnography (PSG) to home sleep apnea testing (HSAT), AI systems are trained to score sleep based on cardiorespiratory signals, to provide sleep stage information even in the absence of neurological signals. This chapter gives an overview of AI-based algorithms for sleep staging using neurological or cardiorespiratory signals, presents comparisons of hypnodensity graphs derived from multiple manual scorings and from AI-based autoscoring, and discusses potential new applications of using the hypnodensity instead of the classical hypnogram for evaluating sleep.