Authors :
Miriam Guendelman, MD, MSc - Ben Gurion university of the Negev, Beer Sheva, Israel; Rotem Vekslar, BSc - BSc student, Ben Gurion University; Oren Shriki, PHD - Head, Dept. of Cognitive and Brain Sciences | Head of the Computational Psychiatry Lab, Cognitive and Brain Sciences, Ben Gurion University
Rationale: The diagnosis and management of epilepsy begin with seizure type classification, based on clinical characteristics and electrographic patterns identified by a trained clinician in surface EEG. One of the main weaknesses of this approach is that it does not consider the electrophysiological dynamics of seizures. A recently presented approach, based on concepts from non-linear dynamics and bifurcation theory, suggests a taxonomy of 16 electrophysiological seizure types, termed
dynamotypes (Saggio et al. 2020; Jirsa et al. 2014). These dynamotypes are defined by the manifestation of the seizure's initiation, evolution, and termination in the brain. A recent study analyzed intracranially recorded data from 120 patients with focal onset seizures to identify seizure onset and offset bifurcations (
Saggio et al. 2020). However, the utility of this classification in surface EEG is still unclear.
Methods: We used non-invasive, surface EEG recordings from 160 patients with focal onset seizures. Applying dimensionality reduction and blind source separation, we identified independent signal components with a prominent seizure initiation and termination. Subsequently, we used these seizure components (SCs) to classify the seizures into different dynamotypes.
Results: We found clear SCs in a significant proportion of the recorded seizures, which allowed us to classify the bifurcation types of seizure onset and offset based on visual inspection. Notably, the bifurcation distribution was similar to the distribution reported in the previous study (
Saggio et al. 2020). In particular, the dominant bifurcations at seizure onset were saddle-node and subcritical Hopf, and the dominant bifurcation at seizure termination was the fold limit cycle. In addition to visual classification, we examined the utility of an automated, rule-based classifier to identify the seizure components and the bifurcation types.
Conclusions: Our study demonstrates that it is possible to identify seizure dynamotypes in surface EEG recordings, complementing the clinical classification in routine patient evaluation. Thus, the combined approach can potentially lead to a more objective seizure classification and better treatment adaptation. Furthermore, precise identification of dynamotypes may also open the door for non-invasive responsive neural stimulation as an antiseizure treatment.
Funding: Please list any funding that was received in support of this abstract.: This research was supported by Israel Science Foundation grant 504/17 to O.S.