Abstracts

Rationale: Wide bandwidth recordings (>100 Hz) have opened new opportunities for understanding the mechanisms underlying epilepsy. In addition to prior observations of pathological, high-frequency oscillations and microseizures in humans, pathological unit activity in animal models of epilepsy has been associated with seizure onset and holds promise for aiding seizure prediction. Recent advances in both acquisition and storage technology now make it feasible to record continuously at sampling frequencies (>20 kHz) sufficient to isolate the action potentials of single neurons ( unit spikes ) from multiple electrodes for days at a time. Methods: Continuous data sampled at 32 kHz were collected from multiple, microelectrodes (diameter ~ 40 m) from patients with epilepsy undergoing evaluation for resective surgery. Unit activity was detected offline by identifying peak voltages greater than 3 standard deviations above the mean (with a minimum peak amplitude of 10 V). These detections were further restricted by matching to a generalized template for action potentials. Periods of movement-related artifact and seizure onset times were identified by visual inspection, and were stored along with unit activity detections to a relational database system. Results: Data were collected from 8 MTLE patients implanted with hybrid depth electrodes, which contained a total of 298 microwires in addition to standard clinical contacts. Over a total of 88.1 hours of recording, a total of 24 spontaneous seizures were recorded. Conclusions: The amount and variability of data recorded from microelectrodes across multiple days provides a novel set of challenges for identifying unit activity compared to more common unit activity studies, in which recordings last a few hours. The ratio of the number of detections observed on an electrode compared to the average RMS of that same electrode proved a useful method for the automated identification of bad (i.e., noisy ) electrodes from more suitable candidates for the detection of unit activity. The identification of nearly simultaneous detections on multiple channels recorded in parallel is also an effective method for removing artifactual detections.