A Wavelet- and Machine Learning-Based Framework for the Automatic Detection of Artefacts in Electromyography REM sleep

Rapid eye movement sleep Without Atonia (RWA) is the polysomnographic hallmark of REM Sleep Behavior Disorder (RBD), manifesting either as elevated background tone or phasic activity. Scoring RWA relies heavily on visual inspection of electromyography (EMG) signals from polysomnography (PSG) recordings. This process is time-consuming and prone to inter-rater variability, particularly due to the presence of artefacts. Currently, no standardized method for artefacts removal is available. This study proposes a Matched-Wavelet approach to characterize the morphology of EMG signals, and a Machine Learning (ML) based framework to identify artefacts from EMG recordings during REM sleep, to facilitate subsequent RWA scoring, by decreasing manual labour. The best models achieved F1 scores of 79.2% and 86.3% in detecting artefacts from background and phasic activity, respectively. These results suggest the feasibility of automatically remove artefacts through a low-computational cost method, leading to improved reliability in RWA assessments.Clinical relevance- The framework provides a robust tool for the assessment of artefacts in EMG recordings, improving the reliability of RWA assessment, and contributing to improved diagnostic accuracy.