A Compact Quiet Sleep Estimator Based on Cardiorespiratory and Video Motion Features for Maturation Analysis in NICU

Area of Science:

• Neonatal Medicine
• Biomedical Engineering
• Sleep Science

Background:

• Monitoring sleep in premature infants is crucial for identifying potential health issues.
• Automated sleep stage detection can improve clinical care and developmental assessments.

Purpose of the Study:

• To develop and validate an automated system for estimating and tracking Quiet Sleep (QS) in newborns.
• To investigate the evolution of QS patterns in relation to post-menstrual age during hospitalization.

Main Methods:

• Utilized ECG, respiration, and video motion features from 33 newborns (including preterm infants).
• Employed feature extraction, selection, and evaluated Random Forest, Logistic Regression, and K-Nearest Neighbors classifiers.
• Developed a robust model with a switching mechanism for scenarios with limited video data.

Main Results:

• Achieved a balanced accuracy of 84.67.5% for QS estimation using a compact and interpretable model.
• Demonstrated an increase in QS duration and mean interval duration with advancing post-menstrual age in preterm infants.
• Validated the model's robustness with a large dataset of 18 newborns (1396.6 hours).

Conclusions:

• The developed automated QS estimation model shows high accuracy and robustness.
• Continuous, non-invasive monitoring of sleep patterns provides valuable insights into preterm infant development.
• Findings support the integration of such technologies in neonatal intensive care units for enhanced patient care.