Machine Learning Automated Analysis Applied to Mandibular Jaw Movements During Sleep: A Window on Polysomnography

Sleep monitoring of mandibular jaw movements (MJM) shows physiological displacement of only a few tenths of a millimeter which is linked to the breathing cycles and controlled by the respiratory centers. This monitoring provides the ability to accurately measure the changes in respiratory effort (RE) across sleep-disordered breathing through a single point of contact sensor placed on the patient’s chin. The inertial unit included in the capturing technology and overnight positional stability of the sensor produce a robust bio-signal. Associations between the MJM bio-signal properties and both physiological and pathological breathing patterns during sleep have been extensively studied. These show a close relationship between the changes in the MJM bio-signal amplitudes with those measured simultaneously by reference signals as a function of the RE level. Specific waveforms, frequencies, and amplitudes of discrete MJM are seen across a variety of breathing disturbances that are recommended to be scored by the American Academy of Sleep Medicine (AASM). In addition to detecting sleep breathing disturbances, this new MJM bio-signal also provides information about rhythmic masticatory muscle activities (RMMA), arousals, sleep/wake states, and sleep stages, enabling accurate calculation of the conventional hourly indices. The MJM bio-signal can be manually interpreted and its automatic analysis using a dedicated machine learning algorithm delivers a comprehensive and clinically informative sleep study report opening a window on the conventional polysomnography.