Automatic detection of rapid eye movements (REMs): A machine learning approach

Area of Science:

• Neuroscience
• Sleep Science
• Biomedical Engineering

Background:

• Rapid eye movements (REMs) are key indicators of REM sleep, crucial for memory consolidation and psychopathology research.
• Current manual REM detection is subjective and labor-intensive, necessitating automated solutions.
• REM density, the number of REMs over time, is a significant research metric.

Purpose of the Study:

• To develop and validate an automated algorithm for detecting rapid eye movements (REMs) during sleep.
• To compare the performance of the automated algorithm against human expert scorers and non-expert raters.
• To establish a reliable and efficient tool for REM density analysis in research.

Main Methods:

• Developed a novel automated REM detection algorithm using extracted features and the AdaBoost classification algorithm.
• Utilized Electrooculogram (EOG) data from the right and left outer canthi (ROC/LOC).
• Evaluated algorithm performance using Recall and Precision metrics, comparing against human expert and non-expert scorers.

Main Results:

• The automated algorithm achieved 78.1% Recall and 82.6% Precision, surpassing average human detection performance (76% Recall, 83% Precision).
• Inter-rater reliability was significantly higher among experts (Cronbach Alpha=0.80) compared to non-experts (Cronbach Alpha=0.65).
• The developed algorithm outperformed all previously published LOC- and ROC-based REM detection methods.

Conclusions:

• The automated REM detection algorithm is a viable and efficient tool for sleep research.
• The algorithm demonstrates reliable performance comparable to human scorers.
• This automated method offers a significant improvement over existing REM detection techniques.