Forecasting Seizure Duration from Neural Connectivity Patterns in a Neural Mass Model

Abstract

Understanding seizure duration is crucial for predicting seizure evolution and developing targeted therapeutic strategies. This study investigates how changes in neural connectivity patterns estimated in a Neural Mass Model (NMM) relate to seizure duration and whether these parameters can be used to forecast seizure duration. We applied a biologically plausible NMM to intracranial EEG recordings from a Tetanus Toxin rat model of epilepsy. Model parameters, representing synaptic connectivity strengths between excitatory and inhibitory neuronal populations across different cortical layers, were estimated using the Unscented Kalman Filter. A Random Forest classifier was trained to predict seizure duration (short vs. long) based on these connectivity parameters. We assessed the classification performance using the receiver operating characteristic (ROC) curve and area under the curve (AUC). Our findings reveal that stronger excitatory-to-excitatory connectivity before seizure onset was associated with longer seizure durations. In contrast, inhibitory-to-inhibitory and excitatory-to-inhibitory connectivity strengths decreased before long seizures. The Random Forest classifier achieved an AUC of 0.91 when using both preictal (1 minute before seizure onset) and early ictal (first 5 seconds of the seizures) parameters, demonstrating excellent predictive power. Classification using only preictal parameters resulted in an AUC of 0.70. Our study highlights the critical role of preictal neural connectivity in determining seizure duration. The findings suggest that network excitability and inhibitory control before seizure onset influence how long seizures persist. These insights provide a foundation for personalized, time-based therapeutic interventions and seizure management strategies, ultimately improving the quality of life of patients with epilepsy.