Seizure Prediction: Improving Prediction Efficacy by Combination of Algorithms
Abstract number :
1.033
Submission category :
Clinical Neurophysiology-Computer Analysis of EEG
Year :
2006
Submission ID :
6167
Source :
www.aesnet.org
Presentation date :
12/1/2006 12:00:00 AM
Published date :
Nov 30, 2006, 06:00 AM
Authors :
1,2Hinnerk Feldwisch genannt Drentrup, 1,2Björn Schelter, 1,2Matthias Winterhalder, 1,3Jakob Nawrath, 1,3Johannes Wohlmuth, 3Armin Brandt, 1,2Jens Timmer, and 2,3Andre
Concerning the assessment of the predictability of epileptic seizures, continuous and reliable analysis of neurophysiological recordings of epilepsy patients is necessary. Recent work has indicated though not unequivocally accepted that methods from non-linear dynamics can detect significant preseizure changes in the EEG at least for some patients. To increase seizure prediction performance we tested whether or not a combination of different seizure prediction methods yields better prediction results., A bivariate phase synchronization index (PSI) and the univariate [ldquo]Dynamical Similarity Index[rdquo] (SIM) were adapted for seizure prediction (Mormann et al. Physica D 2000;144:358-369 and Le Van Quyen et al. NeuroReport 1999;10:2149-2155). The analysis of these prediction methods was based on long-term intracranial EEG data with continuous recordings of 14 patients for up to 14 days (mean 7.5) including 330 seizures. A software framework was developed for an online evaluation of these continuous data. Approaches to combine methods were applied and compared to the individual methods. All results have been validated by a statistical test procedure (Schelter et al, Chaos 2006; 16: 013108)., By evaluating the EEG data continuously and without knowledge of [apos]future[apos] events, a prediction situation analogous to an online application was possible. For a fixed set of parameters, significant prediction results could be observed for four patients by evaluating each method individually, with mean sensitivities of 37% (PSI) and 42% (SIM) for a maximum false prediction rate of 0.15 false predictions per hour. A combination of both methods showed improved results, as significant results could be observed for nine patients, with a mean sensitivity of 48%., Whereas several former studies of seizure prediction are based on short and especially selected EEG recordings, our study was performed in a continuous fashion on long-term data. By application of two individual methods significant results could be observed only for few patients, while a combination of both methods shows significant results for more then half of the patients. These findings represent a next step towards a clinical application of seizure prediction., (Supported by the German Federal Ministry of Education and Research (BMBF grant 01GQ0420) and the German Science Foundation (Ti 315/2-2; Sonderforschungsbereich-TR3).)
Neurophysiology