Clinically localized seizure focus maybe not exactly the position of abating seizures: a computational evidence

Abstract

By modeling the brain as a network, the challenge of abating seizure can be recast as a problem of network control. In the premise of bringing network under control, the minimum number of nodes prerequisite for controlling seizures are thus a natural aim of interest, which is still an outstanding issue. Here, we use the network structural control theory to guide the selection for the optimal control nodes with the aim of fully abating seizures. Firstly, we construct the dynamical complex network of pathological seizure by estimating the synchronicity and directionality of information flows over time between EEG signals from 10 patients with focal epilepsy. Then, based on the controllability and observability principles of complex systems, the minimum key nodes which are effective to fully control the network seizure behaviors are obtained. Results show that the calculated control nodes are distinct with the focus zones from clinic report. This suggests that the full control of epileptic network may not only related to the focus zone, the other non-focus nodes could also play important roles. This finding is validated by using the spatiotemporal neural network model connected with our modeled dynamical adjacent matrix. It successfully reproduces the original EEG signals which can be effectively abated by applying pulse stimulation on the identified key nodes or resecting them, while the partial effects can be obtained when functioning onto the clinically identified focus zones of 60% patients. Interestingly, for another 30% patients lesser nodes than clinic reports are need to fully control seizures. In addition, our work facilitates to identify the evolution paths of information flows, so the non-clinic focus zones identified through controllability principle can be supposed to be the potential seizure foci. In sum, our work propose a general methodology or strategy for seizures focus localizations that could comprehensively consider the time-evolving information flow and the analysis of controllability mechanisms driven by the real seizures data, as well as the computational validation. This may promote to develop the canonical computational framework with the core intent of providing support for clinical treatment decisions.