Abstracts

RATIONALE: We have shown in the past that mesial temporal lobe seizures are preceded by a preictal transition that evolves over minutes to hours (J. Combinatorial optimization, 2001, 5, pp. 9-26; Epilepsia 2001, 42S7, p.41). The transition is detectable by monitoring the maximum Lyapunov exponents (STLmax) of critical cortical sites. In the present study, we present evidence that monitoring the convergence of the dynamical phase of corresponding critical cortical sites also can be used for seizure prediction. We report sensitivity and false positive rate per hour of a seizure warning algorithm based on dynamical phase monitoring (SWAP).
METHODS: Continuous 28- to 32-channel long-term intracranial EEG recordings previously obtained in patients with medically intractable partial seizures were used in this analysis. The method for the estimation of the dynamical phase profiles from the EEG is described in L. D. Iasemidis et al., [dsquote]Phase entrainment and predictability of epileptic seizures[dsquote] in Biocomputing, P. M. Pardalos and J. Principe, Eds., Kluwer Academic Publishers, pp. 59-84, 2002. The phase profiles were estimated per electrode site. Then, instead of the STLmax profiles, the phase profiles were used as an entry to the SWA. Warnings were issued and evaluated, using the same criteria for convergence and length of the time horizon respectively, as in the original SWA.
RESULTS: In 4 patients with a total of 49 seizures (range of 7 to 20 seizures per patient) and 415 hours of EEG (range of 70 to 140 hours of continuous recordings per patient) the overall sensitivity of SWAP was 84.44%, the false prediction rate was 0.205 false predictions per hour, and the average warning time per seizure and patient was 83 minutes. It is noteworthy that the range of warning time across patients was 78 to 99 minutes. The sensitivity of the method subclinical seizures was approximately the same as that for clinical seizures, that is 83.33% versus 85.71% respectively. These results compare well with those of the application of SWA on the same data sets from the same patients with the exception of a higher false positive rate in the case of SWAP. However, in its present form, SWAP is computationally faster than SWA.
CONCLUSIONS: The results of this study show that incorporation of the dynamical phase information of the EEG in a seizure prediction scheme is promising. The speed of the new algorithm is an advantage for its real-time implementation as a monitoring and therapeutic tool, e.g. as part of a monitoring / stimulator / intervention device for epileptic seizures. Considering the results from both SWA and SWAP prediction algorithms, we may conclude that the dynamical progression from the interictal state to a seizure involves both changes in stability (STLmax) and synchronization (phase) of critical cortical sites.
[Supported by: NIH/NINDS NS039687
U.S. Veterans Affairs]