A Bayesian parametric model for quantifying brain maturation from sleep-EEG in the vulnerable newborn baby

Area of Science:

• Neuroscience
• Developmental Pediatrics
• Computational Biology

Background:

• Newborns, especially preterms, can show early sleep maturation deviations via Electroencephalogram (EEG), potentially indicating future cognitive issues.
• Current methods use separate algorithms for sleep staging and brain-age prediction, limiting accuracy across the Postmenstrual Age (PMA) range.
• Mismatches between predicted brain-age and PMA can signal maturational deviations, but current approaches have limitations.

Purpose of the Study:

• To introduce a single, end-to-end Bayesian Parametric Model (BPM) for direct brain-age estimation from EEG.
• To model sleep state maturation within the BPM without needing a separate sleep staging algorithm.
• To improve the accuracy and PMA range of brain-age prediction for early detection of developmental deviations.

Main Methods:

• Developed a Bayesian Parametric Model (BPM) to directly estimate brain-age from EEG data.
• Integrated sleep state maturation modeling within the BPM framework.
• Compared BPM performance against traditional multi-stage sleep staging and brain-age prediction algorithms.

Main Results:

• The BPM achieved a comparable performance (Krippendorff's α = 0.92) to traditional methods.
• The BPM demonstrated superior performance in younger infants (under 30 weeks PMA).
• Preliminary exploration showed BPM's potential in identifying maturational deviations in preterm infants with later abnormal follow-up.

Conclusions:

• The BPM offers a unified and effective approach to estimating brain-age from EEG, outperforming traditional methods at younger ages.
• This model can potentially detect early signs of cognitive problems in newborns, particularly preterms.
• The BPM's ability to model sleep maturation directly enhances its utility for assessing neurodevelopmental trajectories.