Multi-Classification of Epileptic High Frequency Oscillations Using a Time-Frequency Image-Based CNN.

High Frequency Oscillations (HFOs) in intracranial ElectroEncephaloGraphic (iEEG) signals are considered as promising biomarkers for localizing the epileptogenic zone. Visual marking of these particular activities is the typical way not only for the identification of HFOs but also for their discrimination from other transient events such as Interictal Epileptic Spikes (IESs). However, this remains a highly time-consuming process. To cope with this issue, several approaches have been already proposed for an automatic detection of HFOs. Most of these approaches are based on machine learning algorithms where relevant features are to be extracted for efficient classification. Looking for these relevant features is however a challenging task and can be avoided by resorting to deep learning. In this paper, a new method for HFOs multi-classification based on a convolutional neural network (CNN) is proposed. The proposed CNN model is based on Time-Frequency representation of HFOs computed using Stockwell transform. The efficiency of the proposed method is confirmed using real iEEG signals and compared with a supervised machine learning approach based on support vector machine (SVM) as classifier.