Abstracts

Rationale: The multi-electrode, multi-day electroencephalography (EEG) collected from intracranial EEG (iEEG) studies from people with refractory epilepsy has played a vital role in advancing human epilepsy research. Common clinical practice involves acquiring data at a sampling rate below 2 kHz, which is sufficient for manual review of EEGs; however, it may not be sufficient for the development of robust quantitative EEG (qEEG) methods. Additional technical development is necessary to facilitate routine collection of data of a higher resolution than the clinical standard. A high-resolution iEEG repository may enable identification of EEG features that are not traditionally observed and allow researchers to reliably test qEEG methods across numerous datasets.

Methods: We built an acquisition infrastructure for the adult and pediatric Stanford hospitals to securely route high-resolution iEEG data with minimal clinical burden. This allows neuroelectrophysiology from intracranial electrodes implanted for clinical studies to be simultaneously and independently acquired from the clinical and research ports of a clinical acquisition system. Pocket-sized routers were used to deploy an encrypted network tunnel that transmitted 10 kHz iEEG data of up to 256 electrodes from the clinical acquisition system in the patient room to a research computer in the hospital server room. We were able to use this higher quality data repository to propose and evaluate the performance of the inverse compression ratio (ICR), an information theoretic technique that shows promise as a qEEG method for seizure detection.

Results: Since September 2017, all eligible patients undergoing iEEG clinical neuromonitoring studies at the adult and pediatric Stanford hospitals were recruited. To date, over 250 TB (800+ days) of neuroelectrophysiology was collected from over 170 routine iEEG studies across more than 150 participants with equal representation of adults and children. Upon analyzing seizure detection algorithms across 30 participant datasets (15 adults and 15 children, 240+ total seizures), ICR outperformed conventional qEEG methods by leveraging changes in information content between seizure and non-seizure periods.

Conclusions: Epilepsy research is limited by the lack of standardized EEG data repositories that could advance the translation of qEEG methods into clinical practice. Our scalable acquisition infrastructure is a potential solution towards building comprehensive data repositories with consistent, higher fidelity specifications and minimal research hardware presence. Our data repository enabled, to the best of our knowledge, the first clinical study that analyzed intracortical EEG for quantitative methods at scale. This clinical study demonstrated the efficacy of multidimensional estimation techniques like ICR (our proposed qEEG method) and its potential usefulness in other domains of biomedical signal processing.

Funding: Supported by the Stanford University Wu Tsai Neurosciences Institute and a Stanford Bio-X Seed Grant Award IIP9-104