Sleep structure discriminates patients with isolated REM sleep behavior disorder: a deep learning approach

Area of Science:

• Neurology
• Sleep Medicine
• Artificial Intelligence

Background:

• Rapid eye movement (REM) sleep behavior disorder (RBD) is characterized by dream-enactment during REM sleep.
• Isolated RBD (iRBD) serves as an early indicator of neurodegenerative diseases.
• Current diagnosis relies on time-consuming, subjective polysomnography (PSG) visual inspection.

Purpose of the Study:

• To develop and validate a novel, fast, and objective deep learning model for identifying iRBD.
• To assess the utility of sleep structure analysis for iRBD classification.
• To improve diagnostic efficiency and accuracy for iRBD.

Main Methods:

• A deep learning model (ResNet-18) was trained using hypnodensity graphs derived from PSG data.
• The study included 86 iRBD patients and 81 controls.
• Model performance was evaluated on whole-night and segmented sleep data (4h, 2h, 30min).

Main Results:

• The deep learning model achieved a mean macro F1 score of 0.784 per subject using whole-night hypnodensities.
• Performance was superior when utilizing entire-night sleep structure data compared to shorter segments.
• Sleep structure analysis proved significant for iRBD classification.

Conclusions:

• Sleep structure analysis via deep learning offers a promising objective method for iRBD diagnosis.
• The developed model demonstrates potential for faster and more accurate iRBD identification.
• This approach may aid clinicians in the early detection of neurodegenerative conditions.