Abstracts

Rationale: Patterns in scalp EEG during seizure activities have been used to categorize the ictal discharges (Ebersole and Pacacia, 1996; 1997). These patterns are often described with respect to the signal frequencies, waveforms, and its spatio-temporal evolutions during the seizure. Quantitatively, it is commonly accepted that an ictal EEG signal, compared to the other periods, has larger amplitude, slower frequency, and higher rhythmicity. However, these signal features may not be consistent across different types of seizures in scalp EEGs. In this study, we investigated quantitative features of ictal scalp EEG signals. It is hypothesized that there exists a common signal characteristic among types of ictal discharges recorded in scalp EEG. Methods: Scalp EEG segments of 88 seizures recorded from 15 patients with temporal lobe epilepsy undergoing long-term monitoring were analyzed. All seizures were verified and reviewed by electroencephalographers, and were classified into three different categories (type I, II or Other) according to their signal characteristics (Ebersole and Pacacia, Epielpsia 37(4), 1996). As a result, 17 seizures were classified as type I, 20 as type II, and the remaining 50 as type “Other”. Multi-channel EEG segments containing 5-minute signals preceded seizure onsets followed by the entire ictal periods were sampled for analysis. Quantitative features, including signal power (via amplitude deviation, AD), signal maximum frequency (via maximum number of positive zero-crossings: ZCmax), and signal regularity (via pattern-match regularity statistic, PMRS, which is smaller when the signal is more regular; Shiau et al., Epilepsia 45(S7), 2004), were then estimated for each non-overlapping 5.12-sec epoch for each EEG channel. In each individual seizure, only EEG channels exhibited clear ictal discharges were included in the analysis. For each feature, the mean changes of the estimated features from the pre-ictal to ictal epochs were analyzed and compared among seizure categories. Results: Quantitative analysis of EEG showed: (1) in all three types of seizures, mean PMRS values of ictal EEG were significantly (p<0.05) different (reduced) from the 5-min “pre-ictal’ period for all seizures; (2) the proportions of seizures with signal power that differed (increased) significantly (p<0.05) from “pre-ictal” to ictal EEG are 0.78, 0.7, 0.56 for types I, II, and “Other”, respectively; and (3) the proportions of seizures with signal frequency significantly (p<0.05) changed (decreased) from “pre-ictal” to ictal EEG are 0.17, 0.45, and 0.22 for types I, II, and “Other”, respectively. Conclusions: The results from this study indicated that, while changes in signal energy and frequency from pre-ictal to ictal scalp EEG seem to be inconsistent among different ictal types as well as within each type, the increase of signal regularity appears to be a common characteristic across different types of seizures, and it is robust within each seizure type. This suggests that the increase of signal regularity could serve as an important feature in developing a scalp EEG-based automated seizure detection algorithm.