Our first speaker is going to be Dr. Nathaniel Sisterson. He's going to be speaking to us on long-term seizure dynamics detected using a novel analytic platform in responsive neurostimulation for epilepsy. Dr. Sisterson. Great. Hello. My name is Nathan Sisterson, and I am a medical student at the University of Pittsburgh School of Medicine and member of the Brain Modulation Lab run by Dr. Mark Richardson. I will be presenting our evaluation of long-term seizure dynamics as detected using a novel platform for analyzing responsive neurostimulation data. My co-authors and I have no disclosures. Briefly, the human brain data driving the study was collected by the RNS system during the course of clinical care. It is currently the only FDA-approved responsive or closed-loop neurostimulation device implanted in approximately 1,400 patients across the country. The device itself is comprised of two bi-directional leads with two channels each for a total of four and operates at a sampling rate of 250 hertz. The onboard storage is limited and can hold approximately four recordings of 90 seconds at any given time. During our initial review of the RNS system, we were struck by what appeared to be rhythmic fluctuations in the number of events detected at a time scale of weeks to months. We weren't the only ones to make this observation. And in a recent Nature Communications publication titled Multi-Day Rhythms Modulate Seizure Risks in Epilepsy from the UCSF group, they undertook a rigorous analysis showing multi-day or multidian rhythms in intraictal epileptiform activity and its correlation with a seizure risk ratio in a subset of patients. Similarly, we theorized that the underlying brain state may contain biomarkers corresponding to seizure susceptibility and hypothesized that multidian rhythms in background theta power correlate with changes in seizure frequency. To test this hypothesis, we had three main aims. The first was to develop a modular pipeline for preprocessing and analytics of closed-loop data. The second was to elucidate multidian rhythms in background theta power. And finally, to correlate these multidian rhythms with seizure tendency. So to obtain and analyze data, we undertook the development of BRAINSTIM, which stands for Biophysically Rational Analytics for Individualized Neural Stimulation. This modular platform automates the collection and analysis of RNS data and is also used by other projects that perform closed-loop data analysis. Using BRAINSTIM, we extracted data from Neuropace, manually reviewed and marked seizure onsets in each recording, and extrapolated detection activity using a combination of reviewed and extracted data. This extrapolation was necessary to reduce bias in the recordings and event counters and was necessary to acquire the time series data for our interictal epileptiform activity and atrial activity. We then identified recordings which contained only background activity, which we defined as having no stimulation events and no manually marked ictal events. We adjusted each ECOG recording for impedance and performed fast Fourier transforms on a per-recording basis to extract theta power. And finally, we regrouped these results by days post-implant. Overall, BRAINSTIM facilitated the review of nearly 15,000 recordings in 12 patients. We identified a subset of those recordings, 5,000 of which contained only background activity according to our definition. To elucidate multidian rhythms in background theta power, we used a combination of periodograms and autocorrelograms to evaluate the presence of rhythmic activity. We did exclude three patients for which we observed no rhythmic activity. To quantify periodicity, we calculated the peak-to-peak distance of the autocorrelation results. Here we show summary periodogram data grouped by lead implant location. So in blue, we have leads implanted in the neocortex. Red represents leads implanted in heterotopic lesions. And green, the mesial temporal lobe. As you can see here, there is a bimodal distribution of periodicity at one week, which persists at two to three weeks, and again, at four to five weeks. These findings were corroborated by the summary data of our autocorrelograms, which also show peaks at approximately one week, two weeks, a month, and then again at two months. We performed these calculations, again, using the same methodology to evaluate periodicity of interictal and ictal epileptic form activity. And again, interictal activity was defined as the number of events detected per day, whereas ictal activity was defined as the extrapolated number of seizures per day. And we found a mean periodicity of approximately 10 days for both theta power and interictal activity, and a periodicity of approximately 30 days for ictal activity. To evaluate the relationship between seizure tendency and the multidian power rhythms, we took the raw theta power data time series and filtered it using Morlett wavelet transforms. We then obtained the instantaneous phase and assigned seizure time series data to corresponding phase bins. The results are seen here in phase diagrams, in which arrow size or length represents the phase locking value, and the direction shows the theta power phase preference for seizures. So on the left, we see that seizures tend to occur during the first half of the uprising phase of background theta power, as evidenced by the cluster of arrows in the lower left quadrant. On the right, we recapitulated a portion of the UCSF analysis, and similarly found a preference for seizures in the second half of the uprising phase of interictal epileptic form activity, as evidenced by the cluster of arrows in the upper left quadrant. These findings indicate that changes in theta power precede changes in ictal and interictal activity, and may have some predictive value. So in summary, BrainSTEM is a modular preprocessing and analytics platform useful for evaluation of responsive neurostimulation data. Our analysis of chronic human brain recording revealed Maltitian rhythms in background theta power, and we saw a preference for seizures to occur during the rising phase of Maltitian theta power. Further development of these results may be clinically useful for informing modulations of seizure drug and stimulation therapies, and notifying patients of increased risk for seizure. Thank you.

long-term seizure dynamics

responsive neurostimulation

epilepsy

RNS system

theta power