Optimal channel and feature selection for automatic prediction of functional brain age of preterm infant based on EEG

Abstract

Introduction: Approximately 15 million premature infants are born each year, many of whom face risks of neurological impairments. Accurate assessment of brain maturity is crucial for timely intervention and treatment planning. Electroencephalography (EEG) is a noninvasive method commonly used for this purpose. However, using all channels and features for brain maturity assessment can lead to high computational burden and overfitting, which can decrease the performance of the prediction system. Methods: In this study, we propose an automatic prediction framework based on EEG to predict functional brain age (FBA) for assessing brain maturity in preterm infants. To optimize channel selection, we combine Binary Particle Swarm Optimization (BPSO) with Forward Addition (FA) and Backward Elimination (BE) methods. For feature selection, we combine the Pearson Correlation Coefficient (PCC), Recursive Feature Elimination (RFE), and Support Vector Regression (SVR) model. Results: The proposed framework achieved a prediction accuracy of 76.71% within ±1 week and 94.52% within ±2 weeks. Effective channel and feature selection significantly improved model performance while reducing computational costs. Discussion: These results demonstrate that optimizing channel and feature selection can enhance the performance of FBA prediction in preterm infants, offering a more efficient and accurate tool for brain maturity assessment.