Features Extraction on Independent Component Analysis Computed on Interictal Magnetoencephalography to Characterize Epileptic Sources
Aurore Semeux-Bernier, Francesca Bonini, Maria Fratello, Elodie Marie Garnier, Samuel Médina Villalon, Jean‐Michel Badier, Frédéric Richard, Christian Bénar · 2023
For pharmacoresistant patients in epilepsy one solution is to resect the brain area responsible for seizures, the epileptogenic zone (EZ). To define the location and the extension of the EZ, several neurophysiological techniques can be used, either non-invasive (magnetoencephalography, MEG, or electroencephalography, EEG) or invasive (stereoencephalography, SEEG). In particular, MEG has been shown to be a useful tool in presurgical diagnosis, but it allows almost only recording of interictal abnormalities which can bear complex relations to the EZ.We propose here a new approach based on the extraction of features from independent component analysis (ICA) computed on the MEG signals in 7 subjects. For each subject we computed 100 ICs on continuous MEG data. These components were labelled by an expert as epileptic or non-epileptic. We test whether different features could individually discriminate the epileptic components from other non-epileptic signals.We show in this preliminary study that some features were significantly separating epileptic from non-epileptic components. But we observed a strong inter-patient variability in feature performance. In particular, automatic extraction of spikes proved to be a difficult issue.