Dual-Task Graph Neural Network for Joint Seizure Onset Zone Localization and Outcome Prediction using Stereo EEG

Syeda Abeera Amir, Artur Agaronyan, William Gaillard, Chima Oluigbo, Syed Muhammad Anwar · 2025

Read Dual-Task Graph Neural Network for Joint Seizure Onset Zone Localization and Outcome Prediction using Stereo EEG by Syeda Abeera Amir, Artur Agaronyan, William Gaillard, Chima Oluigbo, Syed Muhammad Anwar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-23669.txt.

Introduction

Amir, Agaronyan, Gaillard, Oluigbo, and Anwar present a dual task graph network for epilepsy surgery planning in twenty twenty-five. Finding where seizures start and predicting surgical success are vital for drug resistant epilepsy. Stereo EEG records brain activity from inside the skull with high fidelity. Expert reading of those records is subjective and slow.

Their model works on ten second windows from fifty-one seizures in twenty children. Each window becomes a connectivity graph with rich node features. One task predicts whether the patient will be seizure free after surgery. The other flags the channels in the seizure onset zone. Cross validation gave about eighty-nine percent accuracy for outcome prediction. Onset zone localization reached about ninety-five percent accuracy.

Catalog record

CREATORSyeda Abeera Amir, Artur Agaronyan, William Gaillard, Chima Oluigbo, Syed Muhammad Anwar
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-23669

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dual-Task Graph Neural Network for Joint Seizure Onset Zone Localization and Outcome Prediction using Stereo EEG Syeda Abeera Amir, Artur Agaronyan, William Gaillard, Chima Oluigbo, Syed Muhammad Anwar Abstract Accurately localizing the brain regions that triggers seizures and predicting whether a patient will be seizure-free after surgery are vital for surgical planning and patient management in drug-resistant epilepsy. Stereo-electroencephalography (sEEG) delivers high-fidelity intracranial recordings that enable clinicians to precisely locate epileptogenic networks. However, the clinical identification is subjective and dependent on the expertise of the clinical team. Data driven approaches in this domain are sparse, despite the fact that sEEG offers high temporal-fidelity related to seizure dynamics that can be leveraged using graph structures ideal for imitating brain networks. In this study, we introduce a dual-task graph-neural network (GNN) framework that operates on windowed sEEG recordings to jointly predict seizure-freedom outcomes and identify seizure-onset-zone (SOZ) channels. We assemble non-overlapping 10 second windows from 51 clinical seizures spread across 20 pediatric patients, with sEEG data annotated by clinical experts. For each temporal window we construct a functional connectivity graph via thresholded Pearson correlations and extract rich node features (spectral, statistical, wavelet, Hjorth and local graph features), alongside six global graph descriptors. We optimize a combined cross-entropy loss with a tunable task-weight, and select model hyper-parameters via Optuna. Under window-level 10-fold cross-validation, the model achieves a mean graph-level accuracy of 89.31 ± 0.0976 % for seizure-freedom prediction and a node-level SOZ localization accuracy of 94.72 ± 0.0041 %. For the best performing model, we ran additive and leave-one-out ablation studies to explore feature importance for graph and node-level accuracy. Our results demonstrate that jointly modeling global outcomes and local epileptogenic foci in connectivity graphs can significantly enhance prediction accuracy and interpretability, paving the way for AI-assisted, network-informed epilepsy surgery planning. 1 Introduction Epilepsy is one of the most prevalent neurological disorders globally, affecting over 70 million individuals across all demographics. It is characterized by uncontrolled electric discharges in a group of neurons, which result from abnormal electrical activity in the brain [ 1 ] . The condition often interferes with daily life, impacting cognitive function, emotional health, and social interactions [ 2 , 3 , 4 ] . While anti-seizure medications are effective in managing symptoms in many cases, roughly 30 {\%} of individuals with epilepsy suffer from drug-resistant epilepsy (DRE). In these circumstances, seizures continue despite administering pharmacological intervention [ 5 , 6 ] . For individuals with DRE, epilepsy surgery is one of the most effective long-term solutions, with the potential to significantly reduce seizure frequency or even completely eliminate the occurrence of seizures. However, surgical success is heavily dependent on accurate identification of the seizure-onset zone (SOZ), the brain region primarily responsible for initiating seizures [ 5 ] . For cases where lesions are not identified using magnetic resonance imaging, methods for assessing seizure activity involve EEG-based localization using subdural grids and intracranial electrodes. In recent years, pioneering techniques incorporate high quality stereo electroencephalography (sEEG) as an alternative to time-consuming intracranial recordings that used grids, which may take days to weeks [ 7 ] . While these techniques provide a large amount of precise and high temporal resolution data, effectively analyzing the vast amounts of generated information is an intensive, time consuming, and manual process that only specialized epileptologists can perform. Neuroscience and epilepsy research now recognize seizures as a network-driven phenomena rather than isolated focal events. Seizures are understood as disruptions in complex brain connectivity rather than originating purely from a single site [ 8 , 9 ] . This paradigm shift has fueled interest in utilizing computational models that quantify seizure spread and predict intervention outcomes with greater accuracy than would be possible with manual methods [ 10 , 11 , 12 ] . Among the most promising developments in this field is the application of graph neural networks (GNNs), which represent brain activity as a structured network rather than as discrete signals [ 13 , 14 , 15 ] . GNNs differ from conventional signal-processing methods in their ability to capture dynamic relationships between brain regions. By treating electrodes as nodes and functional interactions as edges, GNNs provide a comprehensive opportunity for biologically informed model analyses of seizure propagation [ 16 ] . This enables simultaneous analysis of localized SOZ activity and broader network dynamics, painting a more comprehensive picture of seizure propagation and genesis. Traditional methods of surgical planning rely on expert judgment informed by visual inspection of intracranial EEG such as sEEG recordings [ 5 , 6 ] . This approach, while effective in many cases, is limited by human interpretation, potential bias, and inconsistencies in data analysis. Computational models powered by GNNs provide an alternative data-driven approach that can support clinicians by offering quantifiable and repeatable predictions as well as provide pioneering insights into seizure dynamics. By integrating machine learning techniques with clinical expertise in the form of training data, epilepsy surgery planning can transition from a subjective practice to a more precise, personalized approach tailored to each patient’s unique seizure profile. By harnessing GNNs to model seizure propagation and intervention effects, we aim to provide clinicians with a powerful tool to model epileptic networks, giving patients a greater chance of achieving long-term seizure control and a better quality of life. Our proposed dual-task GNN framework builds upon this foundation by integrating seizure outcome prediction and SOZ localization into a single computational model. Previous studies have typically approached these tasks independently, but our method recognizes their inherent interdependence [ 17 , 18 ] . By jointly optimizing both objectives, we take advantage of overlapping temporal and spatial seizure characteristics to enhance prediction accuracy and model applicability. The practical implications of our proposed framework could significantly improve clinical decision-making. Our Contributions: Our work provides the following major contributions: • We propose a novel Dual-Task GNN for jointly predicting both the seizure onset zone and the seizure freedom outcomes for pediatric patients disgnosed with drug-resistant epilepsy. • We have curated a unique dataset from pediatric patients undergoing sEEG, featuring electrode placements in deep subcortical regions (such as thalamus). This dataset addresses a significant gap in existing research, which has predominantly focused on adult populations and cortical recordings. • Our extensive analysis include features from multiple domains and we show how graph related features are significant in understanding the brain connectivity dynamics during seizure propagation. Related Works GNNs have been explored in recent years for the prediction of seizures and the classification of outcomes in multiple modalities of brain data. Li et al. (2022) proposed a dynamic graph-generative model for scalp EEG, achieving 91% precision in the classification of multiclass seizure types [ 19 ] . Xiang et al. (2025) used a spatiotemporal GNN focused on synchronization in the CHB-MIT dataset, reporting 98% precision and sensitivity in the prediction of early seizure [ 20 ] . Grattarola et al. (2022) proposed an attention-based GNN using intracranial EEG (iEEG) recordings to localize SOZs by highlighting influential nodes in the brain graph [ 21 ] . Similarly, Nejedly et al. (2025) introduced a GNN-based system that integrated interictal stereo-EEG (sEEG), electrode coordinates, and MRI features to automate SOZ localization in epilepsy surgery planning [ 22 ] . Their

Read the complete text