Abstracts

Rationale: Automated spike detection software has been employed for over 25 years in routine seizure monitoring unit applications. Many improvements have been made over the years but automated spike detectors but the sensitivity of the algorithms compared to a gold standard has not been published. Methods: All patients admitted to our seizure monitoring unit over a two week period were considered. Patients with a confirmed diagnosis of psychogenic non-epileptic spells were excluded. Six hours worth of data from both wakefulness and sleep (1500-1700 and 0200-0600 hours) were evaluated with 5 different sets of spike detection parameters with in the Stellate Harmonie software including various combinations of amplitude thresholds, state-dependent spike detection and advanced artifact rejection. The automated spike detections were compared to visual review of the entire six hour file by an experienced epileptologist. Results: Sensitivity was less than 0.50 in each case where spikes were detected. Advanced artifact rejection raised the false negative rate in each case where spikes were detected. An amplitude threshold of 4 had higher sensitivity but did not produce an unacceptible number of detections to review. Conclusions: Despite improvements and adjustments to parameters, this pilot dated suggests that automated spike detection continues to have supoptimal sensitivity and certain features, such as advanced artifact rejection, may have a detrimental effect on sensitivity.