SourceSpikeNet – A Biophysically Grounded AI Approach to Spike Detection in EEG

ABSRACT: SourceSpikeNet – A biophysically grounded AI approach to spike detection in EEG Epilepsy affects approximately 3.4 million Americans, with 40% experiencing seizures despite medication and 50-60% continuing to have seizures after surgical intervention. Electroencephalogram (EEG) detection of interictal epileptiform discharges (IEDs) is crucial for diagnosis, medication selection, and surgical planning. However, current practice relies on subjective, qualitative identification of a small subset of IEDs, with significant inter-reviewer variability. State-of-the-art automated detectors suffer from high false alarm rates, creating an urgent unmet need for precise, objective IED identification. We propose SourceSpikeNet, a biophysically grounded artificial intelligence approach to IED detection that leverages known neurophysiology and electromagnetic physics to improve spatiotemporal accuracy. Our central hypothesis is that a biophysically informed AI can identify IEDs much more accurately than current methods. In Aim 1, we will characterize IED prototype timeseries patterns from 32,433 expert-annotated IEDs using inverse source modeling and self- supervised machine learning. In Aim 2, we will develop SourceSpikeNet by training a deep learning algorithm on an augmented dataset of 3,200,000 simulated spikes from 1,000 uniformly spaced cortical locations. In Aim 3, we will evaluate SourceSpikeNet's ability to detect all IEDs from 2,484 patients' continuous EEG recordings (3-7 days each) and compare comprehensive IED localization to traditional methodology. We expect that SourceSpikeNet will provide superior spatiotemporal localization of IEDs compared to current approaches, with lower false alarm rates and higher spatial resolution. The successful completion of this project will enhance the accuracy of IEDs as a biomarker for identifying seizure foci, potentially enabling higher surgical success rates and improved outcomes for epilepsy patients. Project Number: 1R21NS142800-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Neurological Disorders and Stroke (NINDS) | Principal Investigator: Daniel Goldenholz | Institution: BETH ISRAEL DEACONESS MEDICAL CENTER, BOSTON, MA | Award Amount: $252,060 | Activity Code: R21 | Study Section: Special Emphasis Panel[ZRG1 NINC-Q (01)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11299843