Abstracts

Rationale: Spike frequency has recently been shown to predict outcome in patients with epilepsy (Neurology 2008;71(6):413-8). Frequent epileptiform discharges also lead to transient cognitive impairment (Semin Pediatr Neurol. 1995;2(4):246-53). The wavelet transform provides information about the frequency components of a given signal as well as the temporal location at which those frequency components occur. We hypothesize that wavelet based spike detection may be a helpful tool in the detection of epileptic spikes on EEG, particularly in patients with regional sleep potentiated spiking (SPS) and electrical status epilepticus in sleep (ESES) where spike quantification is an essential diagnostic measure. Methods: Five 300 sec EEG segments, measured at 256 Hz sampling rate, from patients with regional SPS or with ESES were analyzed. The most prominent data channel in each patient was decomposed using a discrete wavelet transform based algorithm, and the resulting high frequency detail output was used to identify localized intervals of high frequency activity which could correspond to epileptic spikes. Subsequently, the background value of the detail output was taken and times where the amplitude of the detail output was greater than twice the background value were marked as spikes. The spikes marked by the computer program were compared against the spike locations marked by a human reviewer with a computerized algorithm comparing the vicinity of both marks. Results: Five patients aged 3 to 17 years with either regional SPS or ESES presenting with seizures and developmental delay were included (Table 1). Cumulative analysis of the data from these five patients included 1,231 spikes marked by the human reviewer, and 882 of these spikes were also detected by the computer algorithm, yielding a sensitivity of 72%. Overall sensitivity of spike detection varied from a lower limit of 53% to an upper limit of 83%. There were a total of 383,793 points classified as non-spikes by the human reviewer, while the computer algorithm classified 383,299 points as non-spikes, yielding a specificity of 99.9%. In all cases, the specificity remained over 99.7%. Conclusions: A computer algorithm based on wavelet analysis may assist in automatic detection and quantification of interictal epileptiform discharges in patients with SPS and ESES. This may prove to be a novel, cost-effective and more precise method to quantify interictal epileptform discharges and to monitor treatment outcome in patients with epileptic encephalopathies who undergo therapeutic intervention to reduce interictal spikes. Future directions will include further algorithm adjustment in order to yield improved spike detection sensitivity while maintaining a high degree of specificity.