Abstracts

Rationale: In the context of drug resistant focal epilepsy, the Virtual Epileptic Patient (VEP) uses a large scale brain modeling approach to predict the spatio-temporal spreading of seizure activity (Jirsa et al., 2017). The localization of the epileptogenic zones (EZ) can be estimated by adjusting the predictions of the model to the empirical stereo-electroencephalographic (SEEG) recordings of the patient. Precise estimates are crucial as they can be considered by clinicians during the presurgical evaluation of the patient and are indicative of candidate regions for resection. In this study, we used high-resolution neural field model (NFM) simulations to evaluate the robustness and the accuracy of the localization procedure of EZ. These NFM simulations provide us with an evaluable ground truth of different hypothetical scenarios.

Methods: To ensure computational tractability, the localization of EZ from empirical data is currently implemented for a VEP with a generative neural mass model (NMM), in which brain connectivity is limited to a global coupling between discrete brain regions. However, to capture the complex spatio-temporal epileptic patterns present in SEEG recordings, the local propagation of brain activity along the cortical surface needs to be considered. This is achieved with NFM in which, in addition to the global coupling, the local coupling within brain regions contributes to the global brain activity. This spatially continuous formulation also ensures an accurate projection of activity from brain regions to SEEG recording contacts, which is a key element for the inversion of generative models. To assess the quality of the estimation and the effects due to the use of a NMM to predict brain dynamics, the estimation procedure was performed on a series of data simulated with a NFM.

Results: The estimation procedure based on NMM was performed for each dataset simulated with a NFM and different parameter inference techniques were compared. The robustness of the procedure was evaluated based on the number and location of correctly identified EZ, while the accuracy corresponded to the capacity of separating EZ from propagation zones.

For the frequent case of mesial temporal lobe epilepsy, simulated data show very realistic seizure propagation pattern and VEP estimates of EZ were consistent with ground truth (see figure).

Conclusions: Overall, the results demonstrate that, given specific epileptic network spatial configurations, NMM may have limitations to capture some complex seizure patterns present in empirical SEEG data. By considering the contribution of both local and global coupling for the propagation of brain activity, NFM can account for a more detailed description of the brain dynamics, as the ones observed during epileptic seizures. The use of NFM for an accurate localization of EZ therefore represents a significative methodological advance to address in the future.

Funding: Please list any funding that was received in support of this abstract.: This work is funded through the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Specific Grant Agreement No. 945539 (Human Brain Project SGA3).