A Deep Learning Approach for Classifying Rapid Eye Movement Sleep Behavior Disorder Using EEGNet.

Area of Science:

• Neurology
• Sleep Medicine
• Artificial Intelligence in Healthcare

Background:

• Rapid eye movement (REM) sleep behavior disorder (RBD) is a key early indicator of α-synucleinopathies.
• Early diagnosis of RBD is crucial for understanding neurodegenerative disease progression.

Purpose of the Study:

• To develop and evaluate a deep learning-based automated framework for RBD classification.
• To assess the framework's performance on multi-center polysomnography (PSG) data.

Main Methods:

• Analysis of REM electromyogram (EMG) data from 227 PSG recordings across five hospitals.
• Implementation of an automated classification framework using EEGNet, comparing manual and automated sleep staging.
• Stratification of data into RBD, non-RBD, Parkinson's disease with/without RBD, isolated RBD, and healthy control groups.

Main Results:

• The EEGNet classifier achieved an AUC of 0.812 with manual staging, outperforming the U-Sleep automated approach (AUC 0.777).
• Solely using REM EMG data was insufficient for differentiating the four clinical subgroups accurately.
• The model demonstrated reasonable performance in detecting RBD.

Conclusions:

• The EEGNet-based classifier shows promise for reducing clinician workload and supporting RBD diagnostic decisions.
• This multi-center study validates an automated and generalizable framework for RBD detection.
• The findings support broader clinical implementation of automated RBD classification.