Abstracts

Rationale:
There are currently 2 FDA approved implantable devices for electrical stimulation of the brain (ESB) for epilepsy, but only approximately 15% of patients using these devices report extended seizure free periods. The data for tracking seizure outcomes in the pivotal trials is from patient diaries, which are notoriously inaccurate. Thus, evaluating the efficacy of brain stimulation is challenging. The development of ESB devices with both sensing and stimulation capabilities opens the possibility of accurate, automated seizure diaries to guide therapy optimization. Here we report initial efforts to develop automated electronic seizure catalogs. In this study, we use 2 power-in-band (PIB) features with a linear discriminant classifier (LDC) to distinguish between seizure and non-seizure states. This simple algorithm can be implemented on the investigational Medtronic Summit RC+STM.
Method:
We investigated automated seizure detection in 1 human and 1 dog with epilepsy implanted with the investigational Medtronic Summit RC+STM sense and stimulation device with leads in bilateral hippocampus and anterior nucleus of the thalamus (ANT). We collected long-term ambulatory intracranial EEG (iEEG) data and implemented a MATLAB mock-up of the LDC classifier embedded on the RC+STM device. Based on the spectral characteristics of seizures, we targeted low frequency (2-8 Hz) and high frequency (15-45 Hz) PIB features. We used the first 75% of seizures without stimulation from each subject as training data for a simple LDC with the 2 PIB features, and tested it separately on out-of-sample stimulation –off and –on states. We ran the algorithm separately for hippocampus and ANT.
Results:
We selected 4 hours of iEEG from the subjects for analysis, including 18 and 31 seizures from human and canine respectively. When stimulation was off, our algorithm was able to detect seizures with a sensitivity of 100% in human ANT but only 33% in the canine ANT. During 100 Hz stimulation in the ANT, our algorithm was able to detect 100% of human and 75% of canine seizures in the hippocampus. The specificity of our algorithm was 100% in all tests, regardless of stimulation being on or off.
Conclusion:
Our results show that a detection algorithm compact enough to be embedded in an existing implantable device has promising specificity, but detecting seizures during electrical stimulation is more challenging. Implementation of an embedded seizure detection algorithm on implantable devices will be valuable for responsive neurostimulation and optimizing epilepsy management. Significantly, in the short-term, patients could receive treatment that matches their true seizure profile. In the future, we will work to further optimize detectors their performance during a wider range of stimulation frequencies and electrode locations.
Funding:
:NIH UH2/UH3 NS095495: Neurophysiologically Based Brain State Tracking & Modulation in Focal Epilepsy