Abstracts

Rationale: Identification of the epileptogenic network is crucial for planning interventions in patients with drug-resistant focal epilepsy. The epileptogenic networks are defined by the network consisted of brain regions involved in the production of epileptic activities. Many statistical methods have been developed for various techniques including MRI, EEG, Stereo-EEG (SEEG), PET to visualize different aspects of the epileptogenic networks based on structural, functional, electrographic, and metabolic abnormalities. Biomathematical modeling offers methods of causal inference for EZ network estimation by integrating multiple modalities and governing the underlying mechanism beyond our observation. The virtual brain (TVB) (Jirsa et al., 2017) technique provides the models in the whole-brain scale to make such causal inference possible.

Methods: A VEP pipeline identifies the EZ network in epileptic patients, using patient-specific brain network models. Connections between regions are estimated through streamlined tractography from DW-MRI, together with the brain parcellation from T1-MRI. Each node contains a neural mass model (2D Epileptor), to represent neural dynamics in a brain region. We compute the source-to-sensor mapping based on a post-SEEG-implant CT in patients’ T1-MRI space. We extract data feature from empirical SEEG recordings to fit the brain network model. Model inversion use, both the LBFGS and HMC, inside a Bayesian framework to obtain a maximum a posteriori estimate of the joint posterior distribution.

Results: We built a solid workflow to estimate the EZ network, for clinical pre-surgical evaluation. The core is the VEP pipeline (Fig.1) which uses a patient-specific brain network model within a Bayesian framework to estimate the EZ network given individual SEEG, diffusion and structural MRI data. Additionally, the validation modules perform validation of the estimates using different a-priori hypothesis, synthetic data and virtual surgery. This workflow has been tested retrospectively in a cohort of 53 patients. The systematical evaluation shows good precision and acceptable recall, comparing VEP results to the routine clinical EZ prediction. The precision shows the VEP results are highly consistent (0.97) with the resected regions for the seizure-free patients. The precision decreased for the patients that are not seizure-free.

Conclusions: The workflow is currently applied prospectively in a clinical trial called Epinov, aiming for 300 patients. The VEP workflow is an important milestone for personalized epileptogenic network estimation and subsequent surgery strategy planning, considering patient specific data from multiple modalities. Each element uses current state-of-the-art technology, but can easily be extended and improved in the future. It provides a measure to further our understanding and knowledge for future fundamental improvement in both brain modelling and epilepsy studies.

Funding: Please list any funding that was received in support of this abstract.: The European Union’sHorizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreement No. 945539 (Human Brain Project SGA3), and the French National Research Agency (ANR) as part of the second “Investissements d’Avenir” program (ANR-17-RHUS-0004, EPINOV).