| Electroencephalography (EEG) and magnetoencephalography (MEG) are indispensable tools used in the diagnostic and treatment of epilepsy. They measure signals that display events heavily linked to epilepsy: the epileptic spikes. These spikes are used by neurologists to confirm their diagnosis but also to localize the region of the brain that causes the pathology. Since current source localization techniques require a high signal to noise ratio (SNR), it is a common practice to average recordings which are assumed to contain similar events. However, how can we be sure the signals are similar enough to be averaged? The answer is to classify the spikes prior to signal averaging.;Keywords: epileptic spikes, classification, clustering, epilepsy, electroencephalography, magnetoencephalography.;In the present work, we present the conception, methodology, and evaluation of a new classification technique based on the source representation of epileptic spikes. Because the source space is used to classify the spikes, the method is able to separate spikes with similar morphologies but generated by different sources. The performance of this algorithm was evaluated using simulated EEG and MEG signals. The results indicate that the method is able to group spikes with similar source representation even if their morphologies are similar. When applied to real data, the method allowed us to identify new active regions of the brain when compared to traditional analysis. |
