Individual-specific resting-state networks predict language dominance in drug-resistant epilepsy

Identifying language dominance is a crucial step in epilepsy surgery planning. We applied a precision functional brain mapping approach to estimate individual-specific cortical resting-state networks in drug-resistant epilepsy and predict language dominance.

To determine whether individual-specific cortical network topography can predict task-based language dominance in drug-resistant epilepsy.

Retrospective case-control study conducted between January 2024 and August 2025.

Multicentre population-based study including healthy participants from the Human Connectome Project, and participants with drug-resistant epilepsy from the National Institutes of Health (NIH) and the University of Iowa.

Eligible participants had drug-resistant epilepsy defined by International League Against Epilepsy criteria and were undergoing pre-surgical evaluation. All participants underwent neuroimaging, with a subset receiving concurrent intracranial electrical stimulation during fMRI.

Individual-specific cortical network topography and prediction of task functional magnetic resonance imaging language dominance.

Ninety-one participants with drug-resistant epilepsy were included: 61 (67.0%) temporal lobe epilepsy, 29 (31.9%) extra-temporal lobe epilepsy, and 1 (1.1%) undetermined seizure onset zone. The mean age was 33.0 ± 11.4 years and 50 (54.9%) were male. There were 40 healthy participants with a mean age of 29.0 ± 4.0 years, and 16 (40.0%) were male.We developed a multi-session hierarchical Bayesian model (MS-HBM) trained on NIH data to estimate individual-specific networks in drug-resistant epilepsy. MS-HBM trained on epilepsy data outperformed group-average networks or MS-HBM trained on healthy participants and generalized well to an independent dataset. During concurrent intracranial electrical stimulation, cortical activation and deactivation aligned more closely to individual-specific networks than group-average networks. Individual-specific language network topography significantly differed across left (mean lateralization index (LI) = 0.165 ± 0.106; area-under-the-curve (AUC) = 0.82), bilateral (LI = 0.056 ± 0.074; AUC = 0.72), and right (LI = 0.023 ± 0.055; AUC = 0.83) language dominance groups (p = 0.002).

Our model is publicly available (github link), which may be used to predict language dominance from approximately 10 minutes of resting-state fMRI. This provides a practical, non-invasive tool for presurgical evaluation of drug-resistant epilepsy.