ABSTRACT The responsive neurostimulation system (RNS) is the first FDA-approved bi-directional brain-computer interface. Developed to treat drug-resistant epilepsy, RNS is an implanted device that automatically records and detects electrographic seizures, then rapidly delivers electrical stimulation to suppress seizure activity. Although the general therapeutic benefit of RNS is well-established, predicting the magnitude and timing of a potential clinical response for each individual patient is difficult. It may take several months for a patient to report a reliable change in seizure status, during which time the programming clinician has no objective guidance regarding whether or not to adjust settings. Although chronic intracranial EEG recordings obtained by the RNS device provide an ongoing window into the neurophysiological state of a patient’s seizure network, there is little knowledge about how to use these recordings in individual patients. Thus, a critical need exists to develop methods for using a patient’s own data to predict when seizure reduction should be expected or to confirm objectively the presence and maintenance of a clinical response. Using RNS recordings, we recently made the first discovery of putative electrophysiological biomarkers that indicate and potentially predict therapeutic response to therapy in individual patients. By visually inspecting the spectral content of >5000 RNS recordings that captured putative seizures, we identified a distinct category of electrographic seizure pattern modulation (ESPM) that was always present in responders and never present in non-responders. In some cases, these ESPMs were observed in RNS recordings prior to patient-reported seizure reduction, suggesting their potential utilization in predicting therapeutic response. These putative biomarkers, however, cannot be identified using the standard RNS clinical user interface. To overcome these data analytic barriers to therapy optimization, we created a software concept for understanding patient-specific RNS performance using intracranial recordings and interpolation of device-recorded data (BRAINStim). Our proposal adds state-of-the- art expertise in machine learning and neural signal processing to develop technology for ESPM detection, characterization, and validation. In the R61 Phase, using recordings from a cohort of 60 subjects (10 pediatric), we will create tools for automatic detection of ESPMs and perform preliminary biomarker validation according to the following Contexts of Use: 1) prediction biomarkers that signal impending clinical response to RNS, prior to patient-reported seizure improvement, which would prevent premature programming decisions, and 2) response biomarkers that can be used to confirm patient-reported outcomes during stimulation and medication adjustments. In the R33 Phase, we will validate ESPM biomarkers in recordings from an extended cohort of 170 subjects (45 pediatric), to justify the use of ESPMs as RNS biomarkers in routine clinical care and novel clinical trials, which will accelerate and improve seizure outcomes, in both adults and children.

NARRATIVE This project combines state-of-the-art expertise in machine learning and neural signal processing to develop technology for detection, characterization, and validation of electrophysiological biomarkers of clinical improvement in patients receiving responsive neurostimulation for drug-resistant epilepsy.