Multi-feature classifiers for burst detection in single EEG channels from preterm infants.

Area of Science:

• Neonatal neurology
• Computational neuroscience
• Medical signal processing

Background:

• Electroencephalographic (EEG) bursts are crucial for assessing preterm infant maturation and prognostication after perinatal asphyxia.
• Existing automated EEG burst detection algorithms are primarily designed for infants under 35 weeks postmenstrual age (PMA), potentially underperforming in older infants.
• Brain activity evolves rapidly, necessitating specialized algorithms for preterm infants approaching term-equivalent ages (PMA ≥ 36 weeks).

Purpose of the Study:

• To develop and evaluate machine learning-based algorithms for automatic EEG burst detection in preterm infants with PMA ≥ 36 weeks.
• To compare the performance of different classifiers, including logistic regression (LR), linear discriminant analysis (LDA), k-nearest neighbors (kNN), support vector machines (SVM), and thresholding (Th).
• To identify features indicative of brain maturation in preterm infants using single-channel EEG data.

Main Methods:

• Five distinct machine learning classifiers (LR, LDA, kNN, SVM, Th) were implemented for EEG burst detection.
• Classifiers were trained and validated using visually labeled, single-channel EEG recordings from 14 preterm infants (born >28 weeks gestation) aged 36-41 weeks PMA.
• Performance was evaluated using accuracy and Cohen's kappa for inter-rater agreement.

Main Results:

• The kNN, SVM, and LR classifiers achieved approximately 95% accuracy, significantly outperforming thresholding (84%).
• Logistic regression, using only three EEG features, demonstrated the highest agreement with human experts (Cohen's kappa = 0.71).
• Application of the LR classifier to a larger dataset (21 infants ≥ 36 weeks PMA) revealed that longer EEG bursts and shorter inter-burst intervals correlate with higher PMA and weight.

Conclusions:

• Logistic regression-based EEG burst detection is a viable and accurate method for assessing maturation in preterm infants at term-equivalent ages.
• The developed LR model, utilizing single-channel EEG, offers a promising tool for monitoring brain development in clinical or portable devices.
• This approach can aid in understanding neurodevelopmental trajectories in preterm infants nearing term.