Abstracts

RATIONALE: There is accumulated evidence that mesial temporal lobe seizures are preceded by a preictal transition that evolves over minutes to hours. We have previously shown that preictal transitions are detectable through a nonlinear dynamical analysis of EEG signals (In Chaos in the Brain, 2000, pp.112). These transitions can be characterized by convergence of the values of STLmax, a stability measure of EEG signals, among critical cortical sites. We have defined this phenomenon as dynamical entrainment. It would be possible to predict an impending seizure only if the critical cortical sites could be identified far in advance. The first automated seizure warning algorithm (SWA), reported in Epilepsia (2001, 42S7, p.41), predicts the occurrence of an impeding seizure based upon the identification of critical sites from the preictal period of the preceding seizure. In the present study, we sought to investigate the null hypothesis that critical cortical sites identified from the preceding seizure (short system[scquote]s memory) are more helpful in predicting the impending seizure than ones from more seizures in the past (long system[scquote]s memory).
METHODS: Continuous 28- to 32-channel long-term (2.9 ~ 5.8 days) intracranial EEG recordings previously obtained in 4 patients with medically intractable partial seizures were used to test the hypothesis. Four methods for selecting the critical cortical sites were compared. Method 1: critical sites only from the preceding seizure; Method 2: critical sites only from the first recorded seizure; Method 3: n groups of critical sites from each of the preceding m seizures, m = 2 ~ 5; Method 4: n groups of critical sites from preceding m seizures, m = 2 ~ 5, where n was the optimal setting for each patient. The sensitivity (% seizures predicted) and the specificity (FPR-false positive rate per hour) of the SWA for each of the four selection methods were estimated.
RESULTS: In 4 patients, the sensitivity and FPR for method 1 were 80.9% and 0.142 per hour, respectively. For method 2 they were 61.7% and 0.178 per hour. For method 3, the sensitivities were 87.2%, 89.4%, 89.4% and 93.6% for m = 2, 3, 4, 5 respectively and the FPRs were 0.289, 0.407, 0.451 and 0.477 per hour. For method 4, the sensitivities were 66.0%, 66.0%, 68.1% and 68.1% for m = 2, 3, 4, 5 respectively and the FPRs were 0.157, 0.181, 0.203 and 0.193 per hour.
CONCLUSIONS: The results of this study show that the our seizure warning algorithm SWA attains minimum FPR per hour at a sensitivity above 80% when the critical cortical sites are identified from the preceding seizure. Including more seizures does not considerably increase the sensitivity but tremendously increase the FPR. The verification of our null hypothesis supports the selection we made in our previously reported SWA and demonstrates the similarity of the participating critical sites in the preictal transition state between closer than further apart seizures in time.
[Supported by: NIH/NIDS NS039687
U.S. Veterans Affairs]