Authors :
Presenting Author: Sheng H Wang, PhD – University of Helsinki/CEA NeuroSpin/INRIA MIND
Morgane Marzulli, BSC – Université Paris Cité, France; Paul Ferrari, PhD – Michigan State University, USA; Vladislav Myrov, MSC – Aalto University, Finland; Gabriele Arnulfo, PhD – University of Genoa, Italy; Lino Nobili, MD, PhD – University of Genoa, Italy; Satu Palva, PhD – University of Helsinki; Philippe Ciuciu, PhD – CEA/Neurospin & Inria/MIND, France; J Matias Palva, PhD – Aalto University, Finland
Rationale: Localizing the epileptogenic zone (EZ) for surgery has remained a challenging task, as evidenced by uncertain postsurgical seizure freedom (30–80%). Our approach to this problem was motivated by Complex Systems theory and the “Brain Criticality” hypothesis, which posit that brains benefit from operating near a critical point of a phase transition between order and disorder. Super-critical behavior, 9i.e., excessive order and hypersynchrony) characterize epileptic seizures. Here, using both computational modeling and stereo EEG (SEEG) recordings from drug-resistant epilepsy patients, we advanced novel biomarkers for seizure zones (SZ). Moreover, we hypothesized that epileptogenic pathophysiology could be simplified into a low-dimensional problem within the criticality framework.
Methods: Ten minutes of SEEG inter-ictal resting-state data from 64 (29.7±9.5 yo, 29 females) patients were studied. Novel criticality and synchrony features were used to train supervised classifiers and localize the SZ identified by physicians. For generalization of the EZ mechanisms, the features of all SEEG samples were dimension reduced to a handful of eigen-feature coefficients. Two unsupervised classifiers were subsequently employed to delineate sample clusters.
Results: Combining all features yielded the most accurate SZ-classification so that the area under receiver-operating-characteristics curve was up to 0.85 vs 0.6 – 0.7 for individual features alone. Unsupervised classifiers revealed a cohort-level sample cluster that engaged in strong synchrony with local high visitability, high inhibition, and aberrant scaling, which strikingly resembled our model in a high seizure-risk regime. Further investigation discovered a smooth, funnel-shaped surface in a low-dimensional eigen-feature space, wherein the individual position was predictive of individual variability in the accuracy of supervised SZ-classification. Importantly, in a particular area on this surface, samples were associated with higher probability of observing interictal spikes, wherein many nonSZ and SZ were indistinguishable by their eigen-features. As the rising phase of the spikes has been shown to be indicative of incoming seizures, this low-dimensional eigen-feature surface likely characterizes an interictal continuum.
Conclusions: We conclude that EZ dynamics encompass anomalies at both the local and network level. The EZ may comprise both SZ and pathological nonSZ. Our approach paves the way for automatized, personalized, and objective determination of EZ for consideration in surgery planning and prognosis.
Funding:
JMP: the Academy of Finland grant (SA253130, 296304);
SP&JMP: Sigrid Jusélius Foundation grant; GA&LN: NEXTGENERATIONEU (NGEU) by the Ministry of University and Research, National Recovery and Resilience Plan, project MNESYS (PE0000006) – A Multiscale integrated approach to the study of the nervous system in health and disease (DN. 1553 11.10.2022); SHW: the Ella and Georg Ehrnrooth Foundation, the Finnish Cultural Foundation (00220071), and the Sigrid Jusélius Foundation (210527) grant. PC&MM: French ANR DARLING funding (ANR-19-CE48-0002).