Validation of aEEG-CSA Neonatal Seizure Detection Algorithm on Hypothermia Treated Infants with HIE
Objective:
To validate a neonatal seizure detection algorithm that is based on extracted clinical features of the aEEG and CSA on a cohort of cooled neonatal patients with hypoxic ischemic encephalopathy.
Methods:
A seizure detection algorithm was designed using aEEG margin features, CSA features, trained on a public dataset of 79 neonatal EEGs with three supervised machine learning classifiers. It was subsequently tested on an inhouse cohort of 23 neonates with asphyxia whose EEGs were collected during cooling therapy.
Results:
The trained Random Forest, Support Vector Machines and Artificial Neural Network classifiers had an AUC of 0.76, 0.77, and 0.77 and an average accuracy of 0.85, 0.86, and 0.85 respectively. Finally, the average AUC across the 10 seizure patients included was 0.85.
Conclusion:
A neonatal seizure detection algorithm that uses a combination of aEEG and CSA clinical features can capture seizures in HIE patients. Performance across seizure patients is not correlated with seizure duration.
