Detection of fast ripples using Hidden Markov Model

Abstract

Studies show that High frequency oscillations (HFOs) can be used as a reliable biomarker of epileptogenic zone, thus many algorithms have been proposed to detect HFOs. Among the wide variety of HFOs, fast ripples (FRs) are important transient oscillations occurring in the frequency band ranging from 250 Hz to 600 Hz. The automatic detection of FRs can be degenerated by the presence of some "pulse-like" events (commonly, the component of interictal epileptic spikes) associated with an increase of the signal energy in the high frequency bands, exactly as in the case of real FRs. The goal of this study is to propose a new method for automatic detection of fast ripples by using Hidden Markov Model (HMM). This method can separate fast ripples from interictal epileptic spikes and background EEG by classifying each segment of signal in three classes. The sensitivity and specificity show this method is reliable to detect fast ripples and avoids false detections caused by sharp transient events often present in raw signals.