Aurore Semeux--Bernier & Ionuț-Flavius Bratu

We are a duo from the DynaMap team, and in this INS seminar, we will present two complementary perspectives on epilepsy research, both rooted in the interplay between methods and clinical questions.  One of us is investigating the interictal period, applying machine learning to MEG data to automatically detect epileptogenic networks, while the other focuses on the ictal and postictal phases, using information theory to explore language disruptions after seizures.

Machine learning to automatically identify the epileptogenic networks in magnetoencephalography - Aurore Semeux--Bernier

Magnetoencephalography (MEG) provides crucial information for the pre-surgical evaluation of drug-resistant epilepsy, but its analysis remains subjective and tedious. Independent component analysis (ICA) can be used to summarize MEG data and identify regions involved in interictal epileptic networks, but its interpretation requires some expertise. This study aims to take advantage of machine learning to facilitate the classification of independent components (ICs) in MEG. ICA was applied to MEG data from 41 epileptic patients, and features were extracted for each IC. This set of features forms the input to machine learning models. They aim to: (i) distinguish cerebral components from artifacts, (ii) identify epileptic ICs and (iii) evaluate epilepsy-related features.  With classical machine mearning model, we could provide excellent artifacts classification and above chance level performances for the classification of epileptic ICs. Our method could also highlight important biomarkers of interictal epileptic activity such as spectrum, dipolarity, connectivity, kurtosis and regularity measures.

Going beyond the seizure: untangling postictal aphasia and brain language networks using an information theory approach - Ionuț-Flavius Bratu

Patients not only experience seizures, but they are also faced with the ictal aftermath. With deficits ranging from mild fatigue to postictal psychosis and aggression, the postictal period can affect the quality of life more than the seizures themselves. One of the most frequent and disabling postictal impairments is represented by language deficit. The current work proposes a more forensic approach to understanding this kind of deficit and from it, inferences on the physiological organisation of language in the brain. We leveraged the analysis of the ictal and peri-ictal stereo-electroencephalographic (SEEG) signal recorded from drug-resistant epilepsy patients using an information theory approach - permutation entropy. Anatomical-electro-clinical correlations were performed based on the recovery of various language domains that were tested during the postictal period and the recovery of the complexity of the signal from the SEEG sampled brain regions. Beyond the structure-function coupling we could ascertain and add to the hypothesis of the bi-stream organisation of language. Moreover, a clinical tool was developed in the form of a postictal aphasia score for bedside testing.

Aix-Marseille Université
INS - Faculté de Médecine, 27, Boulevard Jean Moulin
Marseille, 13005, France

Salle 52

Zoom: https://univ-amu-fr.zoom.us/j/86932285282?pwd=bjRWQ3ozd0ZiNFJOS2ZWSGlzcDFNdz09