Work

Automated estimation of EEG maturity in preterm neonates and its association with long-term outcome

Year: 2025

Type: article

Abstract: Background: The purpose of this study was to train three end-to-end machine learning models estimating the electroencephalogram (EEG) maturational age (EMA) of preterm infants using EEG data from amplitude-integrated electroencephalography (aEEG) monitors and to investigate the association between the predicted age difference (PAD) and long-term neurodevelopmental outcome.Methods: Three convolutional neural network (CNN) models were trained and evaluated using EEG data from 92 preterm infants with a post menstrual age (PMA) of 23 to 41 weeks, without major neurological conditions and with normal cognitive outcomes based on the Bayley Scale of Infant and Toddler Development (BSID-III) at two years corrected age. PAD scores were compared using a larger dataset of 148 infants with both normal and abnormal outcome.Results: The largest CNN model had the best performance, achieving 84% accuracy in predicting EMA within ±1 week of actual PMA and a mean absolute error of 0·55 weeks. PAD values predicted by this model were significantly lower for infantes with severely abnormal outcome (p<0·001).Conclusion: Our CNN models demonstrated high accuracy in predicting EMA and successfully identified significant associations between PAD and neurodevelopmental outcome. Lower PAD values were closely associated with severely abnormal outcomes, highlighting the model's potential for early detection of neurological risks in preterm infants.Funding: This study was partially supported by the FWF (Dr Gottfried and Dr Vera Weiss Wissenschaftsstiftung), project number: KLI 1016-B.Declaration of Interest: The authors have no conflicts of interest relevant to this article to disclose.Ethical Approval: This study was approved by the local ethics committee (EK Nr: 1814/2018) of the Medical University of Vienna.

Source: Clinical Neurophysiology

DOI: 10.1016/j.clinph.2025.2111432