Abstracts

Rationale: The detection of interictal spikes from electrophysiological data plays an important role in epilepsy research, namely for studying interictal spiking activity in relation to seizures, high frequency oscillations, and other pathological activity. Although algorithms for detecting interictal spikes have been extensively published over the last 30 years, a single reliable method has yet to be clearly established. Most methods rely on a k-sigma rule for signal amplitude to register a spike—that is, peak amplitude exceeding a threshold that equals a certain number of standard deviations, k, above the mean background (non-spike) amplitude—while other methods rely on template matching, which may not generalize well to other datasets due to variation in the morphology of a spike as well as the definition of a spike. For either approach, the reported performance measures—counts of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) or derivations therefrom—are often inconsistent and poor results. Moreover, there are no reports of the receiver operating characteristic (ROC) curve of such detectors or corresponding area underneath the [ROC] curve (AUC), which can describe the performance of a detector for several criteria (values of k). Consequently, we describe a new approach to detect interictal spikes and contrast this approach against three benchmarks with data collected from epileptic rats (using the tetanus toxin model). Methods: For each rat, we record local field potentials from multi-microelectrode arrays that are implanted in the hippocampus, randomly construct a dataset of interictal spikes and background, and compute the AUC for each detector on the same dataset for statistical comparison. Results: For our epileptic rats with interictal spikes (n = 4), a one way repeated measures analysis of variance (ANOVA) finds a significant difference (p<.05, Greenhouse-Geisser correction) in AUC between at least two of the methods and a post-hoc test (three pairwise comparisons) shows that the AUC for the new method (M±S.E. = 0.965±0.026) is larger (M.D.±S.E. > 0.199±0.040) than the other methods (p<0.05, for all comparisons). Conclusions: We determine that the new detector provides a higher quantity and quality of detected interictal spikes than previous detectors, and we now may more accurately investigate the effects of electrical stimulation on interictal spiking in our rodent model of focal epilepsy. For instance, contrasting the microarray recordings before and during 50 Hz electrical stimulation of the hippocampus (six minutes each record) for one of the epileptic rats, we observe a reduction (~64.2%) in interictal spikes during the stimulus. In addition, we note that electrical stimulation of the hippocampus may be a promising technique to reduce epileptic activity in focal epilepsy.