Effect of Detection Parameters on Automated Electroencephalography Spike Detection Sensitivity and False-Positive Rate

Abstract

Most seizure monitoring units use the Gotman algorithm or a variation on it for EEG spike detection, but the effect of various detection parameters on its accuracy has not been well established. The authors report sensitivities and false-positive rates for several different sets of detection parameters. Nine patients were studied. For each patient, 6 hours of EEG data were analyzed using five different sets of spike detection parameters including combinations of amplitude thresholds, state-dependent spike detection and advanced artifact rejection. Automated spike detections were compared with spikes found on visual EEG review. Mean spike detection sensitivities for the different parameter sets ranged from 0.09 to 0.34. The highest sensitivity occurred with an amplitude threshold of 4, state-dependent spike detection turned on and advanced artifact rejection turned off. Mean rates of false-positives ranged from 4.2 to 48.6 per hour. The highest false-positive rate occurred with the same set of detection parameters that produced the highest sensitivity. The sensitivity of spike detection with the Gotman algorithm is relatively low. The data favor using a lower amplitude threshold and not using advanced artifact rejection. The false-positive rate increases with improved sensitivity, but it is still within an acceptable range for clinical application.