458 Novel Hypothesis-Driven Classification of Temporal Lobe Epilepsy Explorations

INTRODUCTION: Surgical intervention for medically-refractory temporal lobe epilepsy (TLE) is predicated on the delineation of the epileptogenic zone. In selected patients, anatomo-electroclinical hypotheses are explored by SEEG methodology. Conventional classification systems have divided TLE into mesial and lateral (or neocortical) epilepsy. However, this classification does not account for the numerous cytoarchitectonic regions and functionally distinct networks housed in the temporal lobe and connected structures. METHODS: We collected demographic data, clinical/epilepsy history, neuroimaging, seizure semiology, the analysis of the multidisciplinary patient team, and electrode coordinates in a cohort of 60 patients with drug-resistant TLE. We analyzed structural and semiological features and validated the classification with a recursive-feature elimination random forest machine learning algorithm. RESULTS: The random forest (500 trees, Gini criterion) classifier over semiological features (90% classification accuracy) and electrode location (97% accuracy) showed strong agreement with the proposed system and identified key predictive features. An MRI lesion was present in 67% of cases but did not associate with classification (p = 0.3) nor seizure outcome (seizure-free 69% vs. not 60%, p = 0.5). Following SEEG, 94% patients underwent targeted surgical intervention (82% resection, 12% neuromodulation). In resection patients, one-year Engel I seizure freedom rate was 73% and did not vary significantly by classification (p = 0.9) CONCLUSIONS: SEEG-based hypotheses for TLE can be classified into distinct subtypes, grounded in the cytoarchitectonic and functional diversity of the temporal lobe. This classification expands beyond the conventional mesial/lateral TLE divide and can drive targeted surgical intervention with high rate of seizure freedom. This approach guides both a theoretical approach to TLE and a practical realization that can be tailored to the individual patient.