Purpose: The current project evaluated the feasibility of providing an algorithm that could warn a patient of a forthcoming seizure based on iEEG recordings. Method: The mean phase coherence (MPC) feature (Mormann F et al. Phys Nonlinear Phenom 2000;3-4:358-369.) was implemented and tested in a rigorously, out-of-sample manner. The MPC-feature is based on the synchronization measure, explained through the analytic signal approach where the Hilbert transform is used to find the instantaneous phase of an arbitrary signal. By a relative comparison between two different iEEG channels the phase synchronization was calculated. The feature was employed on the FSPEEG database containing 21 patients with 4.1 seizures in average (Winterhalder M et al. Epilepsy Behav. 2003;4(3):318-325.) to assess its predictive performance. Results: A sensitivity of 55% and a specificity of 62% were obtained after a unified optimization of threshold value and localization of electrodes for all patients. These results are just better than a random predictor. To improve the results the parameters need to be optimized for each patient individually. Before this can be done, a larger database with more seizures recorded per patient is needed. Conclusion: It was shown that it is possible to anticipate an epileptic seizure to some degree. While the obtained results are still far from clinically applicable, they suggest that by optimization of personal parameters, at least some patients will be able to gain advantage of seizure prediction. The field still needs further investigation though.
|Published - 2009