A Novel NICU Sleep State Stratification: Multiperspective Features, Adaptive Feature Selection and Ensemble Model.

Area of Science:

• Neonatal neuroscience
• Computational neuroscience
• Medical informatics

Background:

• Sleep pattern analysis is vital for assessing neonatal development, especially in premature infants within the neonatal intensive care unit (NICU).
• Current methods for sleep state classification in neonates can be labor-intensive and subjective.
• Objective, automated assessment tools are needed to monitor neurological and physical development in NICU infants.

Purpose of the Study:

• To develop and validate an automated multi-sleep state classification approach for infants using electroencephalography (EEG) data.
• To assess the utility of multiperspective feature extraction and machine learning for analyzing neonatal sleep patterns.
• To improve the accuracy and reliability of sleep state classification in the NICU setting.

Main Methods:

• Utilized electroencephalography (EEG) recordings from 83 neonates across two datasets.
• Employed a six-phase methodology: data collection, annotation, preprocessing (including multi-scale principal component analysis for noise reduction), multi-perspective feature extraction (1,976 features), adaptive feature selection, and classification.
• Extracted features using stationary wavelet transform (SWT), flexible analytical wavelet transform (FAWT), spectral features (alpha, beta, theta, delta waves), and temporal features.

Main Results:

• The automated approach achieved 81.45% accuracy and 71.75% Kappa score with a single EEG channel.
• Performance improved with more channels, reaching 83.71% accuracy and 74.04% Kappa with four channels.
• Using all eight EEG channels yielded the highest performance: 85.62% accuracy and 76.30% Kappa score.
• Leave-one-subject-out cross-validation confirmed the model's reliability.

Conclusions:

• The proposed automated multi-sleep state classification method demonstrates high accuracy and reliability for neonatal EEG data.
• This approach offers a promising, objective tool for monitoring and assessing sleep patterns in NICU infants.
• The findings support the use of advanced signal processing and machine learning for enhanced neonatal developmental assessment.