Pediatric Seizure Forecasting using Nonlinear Features and Gaussian Mixture Hidden Markov Models on Scalp EEG Signals

Abstract

Seizure forecasting systems have been studied in recent years for improving the quality of life for patients with epilepsy and gain further understanding about seizures. A common approach for this is the study of electroencephalography (EEG) recordings, using signal processing techniques and, more recently, machine learning algorithms. A four-stage system is developed for patient-specific seizure prediction; consisting of pre-processing, dimensionality reduction, feature extraction and classification between interictal and preictal EEG signals. A hybrid method using principal component analysis (PCA) and independent component analysis (ICA) is applied for dimensionality reduction. Nonlinear features are selected for the analysis and characterization of the signals. A Hidden Markov Model (HMM) with Gaussian mixture emissions is trained for each type of signal and evaluated as a classifier. A sensitivity of 0.95 and a specificity of 0.86 were achieved.