Abstracts

Rationale: Intracranial interictal spikes are poorly characterized and their clinical and biophysical effects are not well understood. An important step in addressing these concerns is the development of a reliable and efficient method for their detection. Human scoring tends to be error-prone and labor intensive and so computational methods have recently gained favor. Most algorithms have been developed for the detection of interictal spikes in scalp electroencephalography (EEG) with relatively few concerned directly with intracranial EEG (iEEG) [1]. We present an interictal spike detection algorithm developed specifically from considerations of intracranial recordings, validated on a total of 6 hours of iEEG in 5 patients across a total of 1655 intracranial interictal spikes. Methods: Our algorithm utilizes a pattern adapted wavelet method. This produces a wavelet suitable for use in the continuous wavelet transform fashioned from a small set of clinician determined spikes. By thresholding over normalized wavelet coefficients [2], the algorithm can reliably detect spikes in iEEG. Our main dataset was derived from spikes as marked by a single expert clinician across 5 hours of recordings from 5 patients (dataset A - 1387 spikes marked). A further hour of data was collected from one of these patients and scored by a total of four clinicians (dataset B - 268 unique spikes marked). The algorithm was tested against dataset A to probe its performance across electrodes and patients, as well as against dataset B to ascertain its performance relative to multiple readers. Results: Our algorithm reveals operating points which display accuracies of up to 83.3%, and F-measure scores of agreement between the algorithm and dataset A of up to 90.9% [1]. Figure 1 displays performance curves for one active electrode over left inferior temporal cortex from one patient in dataset A. Five different references are considered. The inset shows the mean spike in that electrode referenced to one of four separate references, over the entire hour considered (26 spikes total, 13 in the test set). For this electrode as well as most of the other electrodes considered in our study, false positive detections do not overwhelm true positive instances. In accord with analyses derived from dataset B, the degree of disagreement between individual clinicians and gold-standards derived from overlaps between clinicians, were as large as those between our algorithm and these same gold standards. Conclusions: Our wavelet based intracranial interictal spike detection algorithm displays operating points which make it feasible to reliably institute it across larger datasets. We imagine this tool will aid in the determination of the potential ramifications of these elusive electrophysiological discharges. [1]. Brown III, MW, et al. (2007) Comparison of novel computer detectors and human performance for spike detection in intracranial EEG. Clin Neurophysiol 118:1744 1752. [2]. Latka M, et al. (2003) Wavelet analysis of epileptic spikes. Phys Rev E 67:052902. Support: Charles H. Hood Foundation (FA, WSA), NIH-NINDS K08 (1K08NS066099-01A1) (WSA)