Use of ICA on Interictal SEEG to Define the Epileptogenic Zone
Abstract number :
1.177
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2023
Submission ID :
210
Source :
www.aesnet.org
Presentation date :
12/2/2023 12:00:00 AM
Published date :
Authors :
Presenting Author: Elodie Garnier, ENG – Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
Matteo Demuru, PhD – Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France; Victor López-Madrona, PhD – Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France; Samuel Medina Villalon, ENG – APHM, Timone Hospital, Epileptology and cerebral rythmology, Marseille, France; Fabrice Bartolomei, MD, PhD – APHM, Timone Hospital, Epileptology and cerebral rythmology, Marseille, France; Christian Bénar, PhD – Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
Rationale: The success of epilepsy surgery relies on accurately defining the Epileptogenic Zone (EZ). Stereotactic EEG (SEEG) is a powerful method for this purpose. However, interpreting SEEG is difficult, and can be facilitated by signal quantification (1). Still, extracting useful information from the interictal period is a key challenge. While the value of interictal spike analysis is uncertain (2), interictal connectivity has shown promise in identifying the EZ (3). Interestingly, Independent Component Analysis (ICA) summarize multivariate signals into independent components, separating signals arising from different nodes of the interictal network (4). The aim of our study is to determine whether ICA analysis on SEEG provides a better definition of the EZ than standard analysis.
Methods: Forty seizure-free patients were included in this retrospective study. For each patient, we called Reference EZ, the contacts defined as epileptogenic by clinical interpretation. ICA was computed on the interictal SEEG, and the spike index was calculated on both SEEG and ICA components. We selected the 20 most spiky ICA components and launched connectivity analysis using non-linear correlation h² with delays-based directionality. For the SEEG, a threshold on the spike index delineated the EZ contacts defining the SEEG-based EZ. Similarly, for the ICA, a threshold on the spike index or on the h² values was used to identify the epileptic components. Then, using the ICA topographies, a second threshold was applied to the selected components to find the contacts defining the ICA-based EZ (on spikes only and on spikes + h²). Finally, we compared the SEEG-based EZ and the ICA-based EZ (spikes and spikes + h²) with the Reference EZ using an F1 score. Figure 1 illustrates the pipeline applied to one patient, with the Reference EZ (A), the spike index computed on the SEEG (B) and ICA components (C.1) with the topographies of the selected components (C.2) and the associated h² graph (C.3).
Results: Both ICA methods show statistically significant higher values of F1 compared to the SEEG analysis (SEEGspikes vs ICAspikes: p<
Neurophysiology