Auto-Detection of Hypsarrhythmia EEG in West Syndrome by Dedicated Feature Fusion and Machine Learning

West syndrome (WS) is a neurodevelopmental disorder causing retardation in many patients. Hypsarrhythmia electroencephalography (EEG) and motor spasms are considered as clinical manifestations of WS. Visual inspection of hypsarrhythmia in long-term EEG recordings is time-consuming and unreliable. This study investigates automated hypsarrhythmia diagnosis using machine learning and dedicated feature selection. The 101 WS patients and 155 healthy controls (HCs) were involved. The 15-s representative hypsarrhythmia and nonhypsarrhythmia EEG segments were selected from each WS patient, and a normal EEG segment with the same length was picked from each HC. Amplitude-, spectrum-, entropy-, and correlation-related features were extracted from each EEG segment. Four popular classifiers, i.e., logistic regression (LR), support vector machine (SVM), adaptive boosting (AdaBoost), and K-nearest neighbors (KNNs), were employed to perform three-label classification among hypsarrhythmia, nonhypsarrhythmia, and HC. Dedicated feature selection was implemented to identify an optimal feature subset for effective classification. Accuracy (ACC), sensitivity (SN), specificity (SP), and the area under the receiver operating characteristic curve (AUC) were adopted as performance metrics. AdaBoost produced the best results in most metrics, i.e., 0.975, 0.971, 0.988, and 0997 for ACC, SN, SP, and AUC, respectively. RMS computed in the delta band and entropy features computed in the beta band were identified as the two most relevant features. Our findings provide useful information for clinical hypsarrhythmia diagnosis and may boost the application of machine learning for automated WS diagnosis in clinical practice.