Distinct Synaptic Excitation-Inhibition Mechanisms Underlie Clinically Defined Seizure Onset Patterns

Epileptic seizures exhibit marked phenotypic heterogeneity that reflects distinct underlying network mechanisms, yet these differences are incompletely captured by current clinical classifications. Computational models offer a principled approach to infer latent excitation-inhibition dynamics from intracranial EEG, enabling mechanism-informed seizure characterization. We analyzed 205 seizures from 15 patients with drug-resistant epilepsy from the European Epilepsy Database, covering seven clinically annotated seizure onset patterns. Using the Wendling neural mass model, we fitted five-second iEEG segments by optimizing synaptic excitation and inhibition parameters across four temporal windows spanning 60 s before to 25 s after seizure onset. Model-derived excitation-inhibition changes distinguished seizure types significantly above chance. Classification performance was strongest when combining excitation and inhibition parameters, with peridendritic inhibition being the single most discriminative parameter. Seizure-type-specific signatures were detectable not only during seizure onset and within seizure onset zones, but already during interictal periods and in non-onset channels, indicating that seizure mechanisms are preconfigured tens of seconds before clinical onset and extend beyond focal onset regions. Although all seizure types showed increases in both excitation and inhibition during seizure transition, their timing and magnitude differed systematically. In particular, our study supports and extends prior evidence that high-amplitude slow (HAS) seizures are driven by localized hyperexcitation within the seizure onset zone, whereas low-amplitude fast (LAF) seizures arise from inhibition-driven network mechanisms. Excitation-inhibition signatures were further linked to individual patient characteristics and surgical outcomes, highlighting their potential clinical relevance.