Abstracts

Seizure recognition during polysomnography (PSG) is challenging due to the limited number of channels devoted to EEG. The addition of extended EEG montages during PSG is thought to improve the detection of seizures and epileptic discharges. The purpose of this study is to determine the validity of abbreviated EEG montage seizure detection during PSG. Three blinded electroencephalographers reviewed 116 5-minute digital files containing focal seizures (N=56) or nonepileptic events (sleep-wake transition/artifact; N=60) using an 8-channel montage and an 18-channel montage. Files were rated in each of two ways. First, after deciding whether or not the file included a seizure, a [ldquo]probability of seizure[rdquo] score from 0-100% was assigned reflecting the confidence of the reader that it was a seizure. Second, for those events classi-fied as seizures, readers attempted to localize the epileptic activity as arising from the tempo-ral, frontal or parieto-occipital region. Readers were then asked to provide the probability of correct localization with 0-100% confidence. The data were analyzed using the Adjusted McNemar Test method of Obochuwski. The continuous probability of seizure score was measured using the Receiver Operating Characteristic Curve. Observed agreement among readers was 78% and 84% for the 8-channel and 18-channel montages, respectively, with a higher agreement beyond chance (kappa of 0.52 versus 0.69, respectively) for the 18-channel (P=0.013). Readers were able to distinguish seizures from nonepileptic events better using the 18-channel montage (area under the curve {AUC} =0.91 vs. 0.82 for the 8-channel montage; P=0.004). Although both montages reliably detected nonepileptic events (92% vs. 96% for 18- vs. 8-channels), seizure detection was better using 18 channels (sensitivity=85%) than 8 channels (sensitivity=68%; P [lt] 0.001). Seizures localized to the temporal and parieto-occipital regions were more likely to be identified as seizures and localized correctly while seizures localized to the frontal regions were commonly classified as nonepileptic events and/or had a higher tendency for mislocalization. The likelihood of correct seizure localization was significantly greater using the 18-channel montage, although neither montage did very well. Readers were able to correctly localize 27% of seizures using the 8-channel and 49% of seizures using the 18-channel montage (P [lt] 0.001). Despite the added time, labor and expertise involved in performing and interpreting 18-channel EEG recordings during PSG, we believe that abbreviated EEG montages fail to adequately differentiate epileptic seizures and nonepileptic events arising from sleep. This appears to be particularly true in frontal lobe epilepsy where seizures are most apt to be confused with disorders of arousal.