Abstracts

Rationale: Seizure prediction by analyzing continuous EEG has been a long-term therapeutic goal for closed loop seizure preemption. Yet, current approaches remain suboptimal. Proofs of principle have demonstrated that examining responses to active perturbation rather than passive spontaneous activity might enhance the seizure predictability. In this work, we explored both spontaneous data over long time periods and the use of single pulse electrical stimulation (SPES) to actively probe the ictogenic neural circuitry. We examined neural responses between and before seizures with an eye toward seizure forecasting. Methods: Young male SD rats (2-3 mo, n = 2) were implanted with surface electrodes, EMG pads and intrahippocampal depth electrodes bilaterally. Unilateral intrahippocampal injections of kainic acid were administered to induce chronic epilepsy, while video and EEG recordings were recorded continuously for 3 months. SPES was delivered at 0.5 mA every three to five seconds for active probing. Multichannel EEG data was analyzed by computing standardized features in time and frequency domains and seizure segments were also manually labelled by an expert. We developed pooled and individualized classifiers using support vector machines (SVM). Results: Data corresponding to 39 seizures and 187 seizures were extracted for the two subjects. From the evoked response to SPES (1 s time-window), we identified multiple features across time and frequency domains that significantly change in preictal periods. SVM analysis yielded AUROC of 0.91 and 0.88 on our two individualized datasets, and an AUROC of 0.88 on our pooled dataset. As a control, we computed the same features on the 1 s time-window preceding SPES which reduced the prediction capabilities (AUROC of 0.66 and 0.75 on the individual datasets, AUROC of 0.67 on the pooled dataset). To compare the influence of active probing versus passive recording for seizure prediction, we used a dataset of 1012 seizures from 12 rats previously implanted with the same methodology. Identically trained SVM classifiers yielded AUROC of 0.75 on average on the individualized datasets, and 0.70 on the pooled dataset. Conclusions: We predicted seizures in a rodent epilepsy model with a high degree of sensitivity and specificity. The usage of active neurostimulation was found to provide a significant improvement in the prediction. If confirmed in a larger cohort, these results offer new insights into the mechanisms underlying seizure initiation, and may help improve diagnostic and therapeutic approaches for patients suffering from focal epilepsy. Funding: This work was supported by the NIH (F31NS105161, K24NS088568, T32MH020017, T32GM007753, R01NS062092), HHSN271201600048C (KSW), the Harvard Medical Scientist Training Program (SE), the Paul & Daisy Soros Fellowship (SE), and the Bertarelli Fellowship (NFF).