HRV-based Monitoring of Neonatal Seizures with Machine Learning

Area of Science:

• Biomedical Signal Processing
• Machine Learning Applications
• Neonatal Neurology

Background:

• Neonatal seizures pose significant risks to brain development.
• Accurate and timely seizure detection is crucial for effective intervention.
• Machine learning (ML) offers potential for automated analysis of physiological signals.

Purpose of the Study:

• To benchmark various ML classifiers for neonatal seizure detection.
• To evaluate the efficacy of heart rate variability (HRV) parameters derived from electrocardiogram (ECG) signals.
• To identify optimal feature extraction and selection methods for neonatal seizure monitoring.

Main Methods:

• Extracted time-domain, frequency-domain, and nonlinear-domain HRV parameters from ECG segments (30-180 s).
• Applied Minimum Redundancy Maximum Relevance (mRmR) for feature selection.
• Evaluated ML classifier performance using nested cross-validation on a dataset from 16 newborns with neonatal seizures.

Main Results:

• The Support Vector Machine (SVM) with a linear kernel demonstrated the best performance.
• Optimal results were achieved using HRV parameters from 180-second ECG segments.
• The best SVM model yielded an Area Under the Curve (AUC) of 0.627, 89.7% sensitivity, 34.6% specificity, and a 92.3% good detection rate.

Conclusions:

• ML-based analysis of HRV parameters from ECG is a viable approach for neonatal seizure detection.
• SVM classifiers, particularly with longer HRV segments, show potential for clinical application in neonatal monitoring.
• Further research is warranted to improve specificity and overall diagnostic accuracy.