Automatic and Machine Learning Methods for Detection and Characterization of REM Sleep Behavior Disorder

Rapid Eye Movement (REM) sleep behavior disorder (RBD) is a sleep disorder characterized by the absence of physiological muscle atonia during REM sleep (i.e., REM sleep without atonia—RWA), resulting in the manifestation of dream-related motor behaviors and vocalizations. RWA is the crucial diagnostic criterion for the diagnosis of RBD in polysomnographic (PSG) recordings. In its isolated phenotype (iRBD), which occurs in the absence of accompanying neurological symptoms or signs, RBD represents a precursor to overt alpha-synucleinopathies (i.e., Parkinson’s disease, dementia with Lewy bodies, and Multiple System Atrophy), with a conversion rate of up to 73.5% over 12 years. The international guidelines for assessing RWA encompass visual scoring of polysomnography data, often entailing protracted manual labor. To overcome the limitations of manual RWA quantification, rule-based algorithms have been proposed, though most of them are threshold-based and still require visual PSG inspection. These methods, however, do not tackle the problem of directly identifying patients with RBD. Machine and deep learning models have recently emerged as tools for the automatic detection of RBD, by leveraging various polysomnographic biosignals, as well as other modalities including actigraphy and imaging techniques. These methods facilitate the identification of patients with RBD and further extend their potential to the prediction of the progression from iRBD to overt alpha-synucleinopathies. This chapter provides an exhaustive overview of these models and applications and presents future possibilities and implications for AI in the diagnosis and characterization of RBD.