Abstracts

Rationale: In large part, the focus of the epilepsy community has been to develop approaches to control or eliminate seizures. Recently, in addition to the seizure events themselves, attention has been devoted to interictal spike events (IISEs) prevalent in a large fraction of patients. These IISEs appear as high frequency, high amplitude electrophysiological signals in scalp or intracranial field potential recordings (IFPs). The overarching goal of our work is to further our understanding of the effects of intracranial IISEs through their potential role in the disruption of cognitive function, by combining neurophysiological recordings in the human brain, psychophysical measurements, and computational data analyses. We present here the first phase of our goal, namely the detection of these IISEs, with the view to instituting our algorithm in an automated detection system. Methods: Human scoring of epileptiform artifacts in scalp recordings or in IFPs is time consuming, and performance tends to decline when annotations are performed over long periods of time. We have developed an IFP IISE detection algorithm that uses wavelet analysis as a means for detection. Our algorithm is trained and tested on (separate) data sets of IISEs as delineated by clinicians. The algorithm convolves the IFP with a pattern adapted wavelet which matches the IISEs defined in our training set. We threshold over normalized wavelet coefficients to determine the detections (1). Results: The intracranial IISE detection algorithm has been validated against IISEs as determined by agreement among a majority of (four) clinicians. Figure 1 demonstrates the mean IISE as marked by one clinician in a single electrode in right medial temporal lobe (electrode 11) over the course of one hour of invasive intracranial monitoring. Figure 2 demonstrates performance curves for the algorithm. The training set of the algorithm consisted of half an hour of IFP, with the testing set consisting of the following half hour. The curves in Figure 2 show how the sensitivity of the algorithm scales with the number of false positives per minute, where the algorithm's performance is validated against spike times as determined by the overlap of at least three (out of four total) clinicians (red), or all four clinicians (black). In the latter case, over the half hour of testing (6 clinician determined IISEs total), we achieve a sensitivity of 100% at the cost of ~ 0.43 false positives per minute (~13 false positives total). Conclusions: This pattern adapted wavelet based algorithm has now been tested on a database of recordings. The algorithm will be used in a real-time fashion to gate the presentation of an inventory of images by the occurrence of IISEs - we subsequently aim to measure the potential clinical effects of IISEs on patients with epilepsy during a variety of memory and cognitive tasks, and in a variety of cortical locations. 1. Latka M, et al., Wavelet analysis of epileptic spikes. Phys. Rev. E. 67, 052902 (2003). Support: Charles H. Hood Foundation(FA,WSA),NIH-NINDS KNS066099A(WSA)