Classification of neonatal amplitude-integrated EEG using random forest model with combined feature

Amplitude integrated electroencephalogram (aEEG), a cerebral function monitoring method, is widely used in response to the clinical needs for continuous EEG monitoring. The focus work of this paper is presenting a novel combined feature set of aEEG and applying random forest (RF) method to identify the normal and abnormal aEEG tracing. To that end, a complete experimental evaluation was conducted on 282 aEEG tracing cases (209 normal and 73 abnormal infants). Instead of the traditional aEEG signal processing and diagnosing methods only based on linear features, we considered both statistical and non-linear features. In our experiments, we extracted and combined different types of features for integrated and segmented signals. The experiments examined the RF algorithmic issues including parameter optimization, segmentation of data and imbalanced datasets processing. The performance of the RF was compared to five commonly used classifiers. The result shows that classification accuracy of our method is up to 91.46%. This also indicates our combined feature set is effective for aEEG classification. Besides, the RF-based method can reach exceptional specificity. This novel method to automatically detect aEEG could help medical staff to monitor the progress of infants at all times.