Automatic detection of microsleep episodes with feature-based machine learning.

Area of Science:

• * Neuroscience
• * Computational Neuroscience
• * Sleep Medicine

Background:

• * Microsleep episodes (MSEs) are brief sleep intrusions (<15s) characterized by EEG frequency slowing.
• * Early detection of sleepiness and MSEs is clinically significant for vigilance assessment.
• * The Maintenance of Wakefulness Test (MWT) is a standard laboratory method for evaluating vigilance.

Purpose of the Study:

• * To develop and evaluate automated algorithms for detecting MSEs based on EEG.
• * To assess the performance of machine learning classifiers (SVM, RF, LSTM) for MSE identification.
• * To identify key EEG features predictive of MSEs.

Main Methods:

• * Analysis of MWT recordings from 76 patients, with 53 used for training and 23 for testing.
• * EEG data processed to derive spectral features (delta, theta, alpha, beta power, median frequency) and eye movement quantification.
• * Classification of MSEs using Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory (LSTM) networks.

Main Results:

• * All three algorithms demonstrated high performance in identifying MSEs, evidenced by sensitivity, specificity, precision, accuracy, and Cohen's kappa.
• * Delta power and the theta/(alpha + beta) ratio were identified as crucial features for the RF classifier.
• * Eye movements emerged as a key feature for the LSTM network's MSE detection.

Conclusions:

• * Automated detection of MSEs, based on the study's EEG definition, was successfully achieved.
• * The applied machine learning algorithms (SVM, RF, LSTM) exhibited robust performance.
• * This automated approach shows promise for objective and efficient MSE identification in clinical and practical settings.