Early Prediction of Electrographic Seizures in Neonatal Hypoxic-ischemic Encephalopathy Based on Amplitude-integrated EEG and Clinical Data

We propose a novel machine-learning method for the prediction of later seizure development in infants with hypoxic-ischemic encephalopathy, based on an Adaptive Boosting classifier. The proposed model was supported by only two features extracted from amplitude-integrated electroencephalog-raphy aEEG (after the first 12 hours of recording) and six clinical data available from the newborn's history charts. All features were extracted in the time domain, and don't require performing of standard raw electroencephalography (EEG). The proposed method is verified on datasets recorded from 47 infants with hypoxic-ischemic encephalopathy, gestational age equal to or more than 35 weeks. The achieved performances were Matthews correlation coefficient and area under the receiver-operating characteristic curve, (MCC 0.495, AUC 0.758) respectively, out-performed only available (to the best authors knowledge) the state-of-art ML model also based on clinical and quantitative aEEG (MCC 0.384, AUC 0.720).