Abstracts

Rationale:
Precise localization of the seizure onset zone (SOZ) is imperative for successful surgical treatment of seizures in medically-refractory epilepsy patients. Single-pulse electrical stimulation (SPES) is an innovative method that has been increasingly used to probe seizure networks in patients that undergo intracranial EEG implantation prior to surgery. SPES provides a unique opportunity to investigate cortical excitability as it delivers sub-threshold perturbations to a pathologically unstable network. Therefore, we hypothesized that mathematical models based in dynamical systems theory would provide an efficient computational framework with unique properties that may hold clinical relevance in SOZ localization. We illustrate the use of dynamical systems modeling of SPES data for seizure onset localization in a cohort of epilepsy patients.
Method:
Fifteen epilepsy patients underwent intracranial monitoring and SPES at Johns Hopkins Hospital. For each stimulated electrode pair, we constructed a linear, time-invariant (LTI) state-space model that describes how the activity from each recording contact responds to stimulation. Model accuracy was assessed by calculating the percentage of data points in which the reconstructed data lied within the 95% confidence interval of the mean of individual stimulation responses. Then, the size of the reachable state space, i.e.,  the “largest” network response possible when stimulating the given pair, was computed from model parameters. Finally, a confidence statistic (CS) was calculated to determine how well the size of the reachable state space matched clinical annotations, with high agreement expected for patient cases with low clinical complexity.
Results:
We first confirmed that LTI state-space models can reproduce responses evoked by SPES in iEEG data. After normalizing the amplitudes of the model output, the model reconstructions lied within the 95% confidence interval an average of 97% of the time (+/- 13%). We found that the size of the reachable state space was significantly greater for electrodes within the SOZ and early propagation electrodes than electrodes outside the epileptogenic network (Wilcoxon rank sum test, p< 0.05). We also found that there was more agreement between the size of the reachable state space and clinically-annotated regions in patients of low clinical complexity, as the CS values were significantly higher than those of high clinical complexity (Wilcoxon rank sum test, p< 0.05).
Conclusion:
Analysis of the mathematical properties of brain networks established with SPES and dynamical systems theory provides an efficient framework for characterization of epileptogenicity of the human brain. State-space models capture the unique interaction dynamics between iEEG that occurs when sub-threshold perturbations are delivered to the pathologically connected network, and incorporation of the properties of these models into clinical decision making may improve clinical outcomes.
Funding:
:RJS is funded by the NIH Institutional Research and Academic Career Development Award Program.