Abstracts

Rationale: Human studies of post-traumatic epilepsy injury did not find a consistent relationship between EEG spiking and epileptogenesis, because many subjects developed spikes but no spontaneous seizures. We are testing whether spike frequency decreases after brain injury in subjects that do not develop epilepsy, and whether this has predictive utility. Computer-assisted spike detection algorithms would greatly enhance the feasibility of such studies, but cannot reliably differentiate spikes from many artifacts. We are therefore testing whether down-sampling of EEG can be used to accelerate human spike detection.

Methods: 43 rats experienced varying severities of acute brain injuries at P30 including Rice-Vanucci stroke, penetrating electrode injuries, bleeding, and cortical infection. The animals were implanted with 2-channel electroencephalography (EEG) telemetry devices with epidural wire electrodes over both hemispheres. Animals were monitored 24/7 for an average of 109 days until the battery in the telemetry unit ceased to function. The majority of rats developed electrographic spikes, but not seizures. Methods for EEG down-sampling have been tested on EEG recordings lasting a minimum of 75 days in a test cohort of five animals. Three representative time points are extracted and anonymized per animal: Day 5, Day 40, and Day 75. For each chosen day, 6 out of the 24 hours of EEG are sampled by randomly choosing any 24 of 96 consecutive fifteen-minute time segments. These segments are concatenated, and then further reduced by removing all 2-second EEG epochs that do not contain spike candidates. Spike candidates are defined as both channels containing at least one voltage sample >= 4 σ from the mean voltage of each channel window. A human reader then manually marks interictal spikes.

Results: This analysis pipeline reduces 24 hours of EEG recording to a mean of 0.80 hours (standard deviation of 0.31 hours). We are currently testing whether spike identification can be further optimized by down-sampling the original 24 hour recording to 2 or 3 hours prior to identification of spike candidates, while still maintaining an accurate estimation of spike frequency. We will then implement this method to test whether spike frequency changes over time after brain injury that does not result in epileptogenesis.

Conclusions: Down-sampling of EEG to facilitate manual identification of interictal spikes greatly improves the feasibility of studies of spike frequency vs time after epilepsy, and represents a useful alternative to automated spike detection.

Funding: Please list any funding that was received in support of this abstract.: Supported by NINDS 5R01NS086364 and DOD / CURE W81XWH-15-2-0069.