Validation of a new data-driven automated algorithm for muscular activity detection in REM sleep behavior disorder

Area of Science:

• Neurology
• Sleep Medicine
• Biomedical Engineering

Background:

• Diagnosing REM sleep behavior disorder (RBD) requires documenting REM sleep without atonia.
• Existing automated indices like RAI, FRI, and KEI show moderate performance in identifying atonia during REM sleep.

Purpose of the Study:

• To develop and validate a novel automated, data-driven method for detecting movements in chin and tibialis electromyographic (EMG) signals during sleep.
• To improve the accuracy and specificity of differentiating between healthy controls, RBD patients, and periodic limb movement disorder (PLMD) patients.

Main Methods:

• Utilized sleep data from 27 healthy controls, 29 RBD patients, and 36 PLMD patients.
• Developed a probabilistic model of atonia during REM sleep and identified movements as EMG signals with low atonia likelihood.
• Combined three optimized classifiers using movement percentages and median inter-movement distance during REM and NREM sleep in a 5-fold cross-validation.

Main Results:

• The proposed method achieved average accuracies of 70.8% (REM and NREM) and 61.9% (REM only).
• Accuracies remained robust at 64.2% and 59.8% after excluding apnea and arousal-related movements.
• The method demonstrated superior performance compared to RAI, FRI, and KEI, particularly in classifying RBD.

Conclusions:

• The new automated method offers higher performance in distinguishing between controls, RBD, and PLMD patients.
• Excluding apnea and arousal-related movements is unnecessary for accurate classification.
• Considering muscular activity from both REM and NREM sleep enhances the identification of RBD patients.