Detection of Brain Interictal Epileptiform Discharges from Intracranial EEG by Exploiting their Morphology in the Tensor Structure

Abstract

Detection of interictal epileptiform discharges (IEDs) from EEG signals is the mainstay of diagnosis of epilepsy. The diversity in IED morphologies and their weakness deteriorate the detection performance particularly when the IEDs of different subjects are combined for training. Here, we propose an IED detection system based on tensor factorization in which IEDs with similar morphology are concatenated into the same slice of a tensor. Applying the proposed method to the intracranial EEG 92.9% accuracy has been achieved. This shows that incorporating IED shape diversity into tensor factorization considerably improves the results.