Automatic seizure detection and seizure pattern morphology

Abstract

Objective: We studied the influence of seizure pattern morphology on detection rate and detection delay of an automatic seizure detection system. We correlated seizure pattern morphology with seizure onset zone and assessed the influence of seizure onset zone on the performance of the seizure detection system.

Methods: We analyzed 10.000 hours of EEG in 129 patients, 193 seizures in 67 patients were included in the final analysis. Seizure pattern morphologies were classified as rhythmic activity (alpha, theta and delta), paroxysmal fast activity, suppression of activity, repetitive epileptiform and arrhythmic activity. The seizure detection system EpiScan was compared with visual analysis.

Results: Detection rates were significantly higher for rhythmic and repetitive epileptiform activities than for paroxysmal fast activity. Seizure patterns significantly correlated with seizure onset zone. Detection rate was significantly higher in temporal lobe (TL) seizures than in frontal lobe (FL) seizures. Detection delay tended to be shorter in seizures with rhythmic alpha or theta activity. TL seizures were significantly more often detected within 10 seconds than FL seizures.

Conclusions: Seizure morphology is critical for optimization of automatic seizure detection algorithms.

Significance: This study is unique in exploring the influence of seizure pattern morphology on automatic seizure detection and can help future research on seizure detection in epilepsy.

Keywords: Automatic seizure detection; Detection delay; Detection rate; EEG seizure pattern; Seizure onset zone.