Graph neural networks for enhancedepileptogenic

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216224%3A14110%2F25%3A00142759" target="_blank" >RIV/00216224:14110/25:00142759 - isvavai.cz</a>

Result on the web

—

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Graph neural networks for enhancedepileptogenic

Original language description

Background and aims: Epilepsy surgery often fails due to inaccurate localization of the epileptogenic zone (EZ). This study introduces a new method using Graph Neural Networks (GNNs) to analyze interictal biomarkers like epileptiform discharges (IED) and high- frequency oscillations (HFO). By mapping these features onto a graph that reflects each patient's unique electrode implantation topology, the method aimed to more accurately depict the dynamics of epileptic networks. Methods: A GNN model was developed to pinpoint the EZ, identified as contacts removed during effective epilepsy surgery. The model underwent training and validation through leave- one- patient- out cross- validation involving 31seizure- free patients. Interictal biomarkers were detected across 30 minutes of NREM sleep SEEG, then encoded into a graph structure for each patient as node features. In the graph structure, electrode contacts within 8 mm were connected by edges weighted by the Euclidean distance. Benchmark models, LR (Logictic Regression) and SVM (Support Vector Machines), analyzed the same features without considering implantation topology. Results: The GNN model outperformed LR and SVM in terms of median Area Under the Receiver Operating Characteristic (AUROC) and Area Under the Precision- Recall Curve (AUPRC). Specifically, GNN achieved an AUROC of 0.93 and an AUPRC of 0.69, whereas LR and SVM posted lower scores, nevertheless there were no statistically significant differences between GNN and LR. Conclusion: The GNN model outperformed traditional methods like SVM in modeling SEEG data as graphs, incorporating electrode implantation topology. This approach suggests that acknowledging spatial relationships between electrode contacts, typically overlooked by conventional methods, can significantly enhance localization precision.

Czech name

—

Czech description

—

Type

O - Miscellaneous

CEP classification

—

OECD FORD branch

30210 - Clinical neurology

Project

<a href="/en/project/LX22NPO5107" target="_blank" >LX22NPO5107: National institute for Neurological Research</a><br>

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů