Abstracts

Rationale: Medically-refractory epilepsy patients admitted for intracranial monitoring often stay in the hospital for days to weeks to capture as many seizures as possible to create a surgical resection plan. Patients are weaned off their anti-epileptic medications, sometimes deprived of sleep, or presented with triggering stimuli to invite seizures to occur during iEEG recording. Despite this exasperating and costly process, some patients have an insufficient number of seizures, and are subsequently sent home with an incomplete seizure onset zone (SOZ) localization. We hypothesized that actively stimulating the patient’s native seizure network, which we uncover through transfer function modeling of evoked responses from single-pulse electrical stimulation (SPES), would accurately localize the seizure onset zone, removing the need to record passive seizures.

Methods: We gathered intracranial EEG data from 18 patients that underwent SPES at the Johns Hopkins Hospital. We built stable, discrete, linear time invariant (LTI) transfer function models for each stimulating contact, which resulted in a vector of transfer functions for one electrode stimulation pair. These transfer function models represent the input-output behavior of evoked responses under SPES, from which the magnitude, as computed by the norm of the vector of transfer functions, versus stimulation frequency plots can be computed. From the magnitude versus frequency plots, parameters such as the peak gain, peak frequency, DC gain, AUC, and roll-off were computed. We built a logistic regression model with the difference between hypothesized SOZ regions and non-SOZ regions as features to distinguish patients with successful versus failed surgical outcomes.

Results: We found that the inclusion of a custom parameter, the peak-to-width ratio, the DC gain, the peak gain, and peak frequency as features into the logistic regression model provided the greatest separation of successful outcomes (defined as ILAE Outcome Scores 1 and 2) when compared to failed outcomes (ILAE Outcome Scores 3-6). The area under the curve of the receiver operator characteristic was 0.81 for this model.

Conclusions: We used a computational model of SPES-evoked responses to identify the seizure onset regions in 18 patients with focal, medically-refractory epilepsy and were able to successfully predict surgical outcome with an AUC of 0.81. These results indicate that a transfer function model can accurately identify regions of seizure onset purely with stimulation data. Utilization of this algorithm in routine clinical care may remove the need to passively record multiple seizures in the Epilepsy Monitoring Unit, drastically reducing the patient’s length of stay and decreasing the financial burden on the healthcare system.

Funding: Please list any funding that was received in support of this abstract.: RJS is funded by the NIH Institutional Research and Academic Career Development Award (IRACDA) Program.