Neonatal Seizure Detection Using Combined aEEG and Compressed Spectral Array Features: A Machine-Learning...

Area of Science:

• Biomedical Engineering
• Clinical Neuroscience
• Machine Learning in Medicine

Background:

• Neonatal seizures pose a significant risk to infant brain development.
• Accurate and timely seizure detection is crucial for effective treatment.
• Current methods for neonatal seizure detection can be complex and resource-intensive.

Purpose of the Study:

• To develop a clinically translatable algorithm for neonatal seizure detection.
• To utilize amplitude-integrated electroencephalography (aEEG) and compressed spectral array (CSA) data.
• To integrate machine learning for enhanced seizure identification.

Main Methods:

• Extracted aEEG and CSA features from centroparietal electrodes of neonatal EEGs.
• Utilized a public dataset of annotated neonatal electroencephalograms (EEGs).
• Trained and tested Random Forest (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN) classifiers.

Main Results:

• RF, SVM, and ANN classifiers achieved areas under the curve (AUC) of 0.80, 0.69, and 0.79 for seizure detection.
• Average accuracies for seizure and non-seizure periods were high (0.90-0.92).
• Higher median accuracy was observed in non-hypoxic-ischemic encephalopathy (HIE) patients.

Conclusions:

• A clinically interpretable aEEG-CSA algorithm is feasible for neonatal seizure detection.
• The algorithm effectively extracts standard EEG features coupled with supervised machine learning.
• This approach offers a promising tool for improving neonatal seizure diagnosis.