Exploratory prediction of surgical outcomes in MRI-negative epilepsy using preoperative individualized morphometric networks

To explore whether preoperative individualized morphometric networks derived from structural MRI may provide discriminative information for predicting surgical outcomes in magnetic resonance imaging (MRI)-negative epilepsy.

Thirty-one MRI-negative epilepsy patients who underwent epilepsy surgery were retrospectively included (17 seizure-free and 14 with seizure-recurrence) in this exploratory study. Individualized gray matter morphometric networks were constructed from preoperative T1-weighted MRI based on cortical thickness distribution features. Network-based statistics were used to identify outcome-related subnetworks. Graph-theoretical features extracted from these subnetworks were classified using a random vector functional link (RVFL) network. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Graph-theoretical analysis revealed differences in network properties between seizure-free and seizure-recurrence groups, particularly in degree centrality and nodal clustering coefficient of the right orbital middle frontal gyrus, left inferior occipital gyrus, and left fusiform gyrus (p < .05). The combined graph-theoretical model demonstrated the highest discriminative performance among the evaluated imaging models (AUC = .876) and outperformed the clinical-only benchmark model (AUC = .504), although performance estimates showed variability across repeated splits.

This exploratory study suggests that individualized morphometric network features may contain information related to postoperative outcomes in MRI-negative epilepsy. However, these findings should be interpreted as preliminary and require confirmation in larger multicenter studies.