Inception Loops for Epilepsy

Epilepsy is a debilitating and life-threatening condition, affecting 1% of the US population. A significant portion of these patients have seizures that do not respond to medication therapy. Neurostimulation is often an effective treatment method for these patients, but only in an adjuvant capacity. The seizure reduction experienced by patients is far from curative, and stimulation treatments rarely result in seizure freedom. By contrast, resective surgery, the gold standard in care for drug-resistant epilepsy, reliably results in sustained seizure freedom and an improved quality of life. However, it is important to note that surgery, despite its potential benefits, is a highly invasive procedure that carries certain risks and adverse effects. Moreover, its major limitation is that its use is restricted to brain regions that can be removed without causing a loss of essential neurological function. We believe that the limited effectiveness of current neurostimulation devices can be attributed to the crudeness of stimulation they provide. The brain, being a complex and non-linear system, requires a modulatory approach that matches its intricate nature when targeting networks involved in epilepsy. Achieving optimal control over these networks is likely to necessitate a complex approach. Therefore, we hypothesize that the generation of more complex patterns will allow for better engagement and modulation of these networks. Here we present a novel platform that offers an unprecedented ability to optimize the design of pulse sequences with complex spatiotemporal relationships for the control of epileptic activity. By combining a cutting-edge method for designing high-entropy stimuli and deep learning, we will demonstrate the ability to infer optimized stimulation for seizure control in a rat model. Our paradigm is based on adapting our Inception Loops paradigm, a deep learning framework for solving high-dimensional, non-linear optimization problems in neuroscience, for the pur- pose of constructing multi-patterned stimuli that are optimized to control epileptic activity. The first step involves building data-driven neural predictive models of targeted brain areas, taking ongoing brain activity and stochastic, high-entropy electrical stimulation patterns as input to predict neural activity. Then, in silico optimization identifies those dynamic stimulation patterns across many stimulation channels that suppress epileptic activity, which are then verified in vivo. The study detailed below will investigate the optimization of multi-pattern stimuli in two complementary rodent models. Finally, we will perform a first-in-human study to investigate the ability of multi-pattern stimulation to con- trol interictal epileptiform discharges as a seizure likelihood proxy in patients undergoing intracranial evaluation. Project Number: 1F31NS143314-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Neurological Disorders and Stroke (NINDS) | Principal Investigator: Paul Steffan | Institution: BAYLOR COLLEGE OF MEDICINE, HOUSTON, TX | Award Amount: $50,114 | Activity Code: F31 | Study Section: Special Emphasis Panel[ZRG1 F03B-W (20)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11317511