Rapid eye movement sleep behavior disorder as an outlier detection problem

Area of Science:

• Neurology
• Sleep Medicine
• Biomedical Engineering

Background:

• Idiopathic REM sleep behavior disorder (iRBD) is a significant early indicator of Parkinson's disease.
• Current diagnostic methods for iRBD rely on subjective interpretation, necessitating objective, quantitative approaches.
• REM sleep without atonia is a key characteristic of iRBD.

Purpose of the Study:

• To propose and validate a semiautomatic algorithm for the early detection of Parkinson's disease via iRBD identification.
• To differentiate normal REM sleep from REM sleep without atonia using muscle activity analysis.
• To establish objective criteria for iRBD diagnosis.

Main Methods:

• Employed a semiautomatic algorithm treating muscle activity as an outlier detection problem.
• Utilized surface electromyographic (sEMG) channels, including electrooculography (EOG), across three subject groups: healthy controls, iRBD, and periodic limb movement disorder (PLMD).
• A one-class support vector machine classifier computed a muscle activity score from manual scoring of REM sleep.

Main Results:

• The algorithm achieved high accuracy (0.993 AUC) in distinguishing iRBD subjects from controls and PLMD subjects when combining anterior tibialis and submentalis muscle activity.
• The developed algorithm demonstrated excellent performance on unseen subjects, correctly separating all cases.
• Electrooculography (EOG) channels proved more discriminative for detecting REM sleep without atonia than the traditional submentalis channel.

Conclusions:

• iRBD detection can be effectively framed as an outlier detection problem.
• EOG channels offer a more discriminative measure for REM sleep without atonia compared to submentalis.
• Arousals and PLMD had minimal impact on muscle activity quantification; further analysis during non-REM sleep may enhance diagnostic separation.