Stacked Sparse Autoencoder based Automatic Detection of Ripples and Fast Ripples in Epilepsy

High frequency oscillations (HFOs) have been acknowledged as a putative biomarker of epileptic seizure onset zones (SOZs). Accurate detection of HFOs is significant for the preoperative localization of epileptic SOZs. In this paper, a new method is proposed to automatically detect ripples (Rs) and fast ripples (FRs) from intracranial electroencephalography (iEEG) in epilepsy. A moving-window technique is utilized to segment the filtered signals which are obtained by filtering the raw iEEG signals using two Chebyshev band-pass filters. Two stacked sparse autoencoder (SSAE) models are proposed to automatically detect Rs and FRs, respectively. By optimizing the parameters of the two SSAE models, our method yields higher sensitivity (88.9±2.4% for Rs and 83.2±2.5% for FRs) and higher specificity (92.3±2.8% for Rs and 86.1±2.8% for FRs) than other three methods do.