A Data-Driven Approach for Linking Epileptic Networks and Cognitive Profiles Using Stereo-EEG

ObjectiveNeuropsychological assessment plays an important role in localizing epileptogenic regions during presurgical evaluation. However, its diagnostic potential is constrained by reliance on syndromic models of epilepsy. A network-grounded approach may provide higher resolution structure-function relationships, yet in vivo evidence linking epileptogenic networks to cognitive deficits remains limited. Here, we developed a network-based framework to test the hypothesis that patterns of sublobar epileptogenicity shape distinct cognitive profiles. MethodsRetrospective cohort study of 42 drug-resistant focal epilepsy patients undergoing stereo-EEG (SEEG) and pre-implantation neuropsychological assessment (16 indices). Epileptic networks were quantified using a composite SEEG-derived epileptogenicity metric ( EzPz score) providing a continuous measure of regional epileptogenicity. Sublobar EzPz values and neuropsychological z-scores were analyzed by Pearson correlations, PCA, and hierarchical clustering to derive network subtypes and domain-specific cognitive associations. ResultsMean age 34.9 years and 79% MRI-negative. Significant negative pairwise correlations were seen: dominant temporal with language, non-dominant mesiotemporal with visual memory; and non-dominant frontal with attention and visuospatial. Exploratory PCA/clustering identified nine network configurations with associated cognitive profiles. For example, dominant mesiolateral temporal configuration (86% MRI-negative): naming impairment with preserved verbal memory; non-dominant frontotemporal: severe executive and visual memory impairment; bimesiolateral temporal: severe language deficits with executive and visuoconstructional impairment. InterpretationEach of our nine network configurations were associated with a cognitive profile shaped by sublobar epileptogenic distribution, lateralisation, and network size. These findings support a shift from syndromic to network-based interpretation of neuropsychological data. Our framework may enhance the diagnostic potential of neuropsychological assessment in SEEG hypothesis generation and surgical planning.