Abstracts

Rationale: The efficacy of vagus nerve stimulation (VNS) for the treatment of medically intractable epilepsy varies substantially among patients. The current method of determining optimal stimulation parameters relies on a trial-and-error strategy, taking into account the clinical response and adverse effects. Given the number of adjustable VNS parameters, determining the optimal settings for each patient may take months, putting patients at increased risk of recurrent seizures. Therefore, there is a great need to identify a reliable physiological marker that can help optimize the stimulation parameters in a relatively short period of time. This preliminary study applied a data mining approach to investigate a relationship between EEG patterns and VNS parameters.Methods: Five patients with previously implanted VNS as adjunct treatment for medically intractable epilepsy were studied. Twenty-four hours of scalp-EEG were recorded, while VNS parameters were kept the same as the Normal Mode for each individual subject. The EEG signal was analyzed using support vector machine (SVM) classification. This study quantifies the time evolution of the multi-channel EEG signal using accuracy of SVM separation between a reference EEG segment and an EEG segment occurring at a different time. The SVM was trained to separate a 10 second reference EEG segment (5120 points per channel at 512 Hz for 25 channels) occurring during each stimulation from all successive 10 second EEG segments prior to the next stimulation. This procedure was repeated for all stimulations occurring within the 24 hour EEG recordings. The mean SVM separation accuracy over the duration of the EEG recording was plotted against each VNS stimulation parameter for each patient.Results: The results suggest that the SVM measure of EEG time evolution demonstrates a correlation with the VNS pulse width and stimulation frequency parameters. The patient with the greatest mean value of multi-channel EEG time evolution is associated with the highest VNS pulse width, the highest VNS stimulation frequency, and the lowest monthly seizure rate. The patient with the lowest mean value of multi-channel EEG time evolution is associated with the lowest pulse width, lowest stimulation frequency, and greatest monthly seizure rate.Conclusions: The findings of this preliminary study suggest a correlation of the SVM measure of multi-channel EEG time evolution with vagus nerve stimulation parameters. We propose that VNS may mimic a theorized effect of a seizure, by “resetting” the brain from an unfavorable preictal state to a more favorable interictal state. The physiological effect of such action can be measured using quantitative EEG analysis. The SVM method may be a good candidate for the development of a physiological marker that would help optimize VNS parameters for individual patients in a relatively short period of time, thus reducing the cost and the risk of recurrent seizures. Future studies should compare EEG segments prior to and after VNS implantation, and analyze the EEG evolution of the signal during the ramp-up period, as VNS parameters are gradually adjusted.