Can Your Prediction Algorithm Beat a Random Predictor?

摘要

To date, approximately one in four epilepsy patients cannot be treated successfullyby common therapeutic strategies, so they continue to suffer from unforeseenseizures. A precise prediction at a sufficiently early stage before seizure onsetwould offer new therapeutic options such as warning devices or even seizure-prevention techniques, e.g., by applying electric stimuli [1]. For this purpose,several time series analysis techniques based on the theory of linear and nonlineardynamics have been applied to intracranial and scalp EEG data. For an overview ofthese studies see, e.g., [2, 3]. Significant changes in the EEG dynamics in a rangefrom seconds up to hours in advance of seizure onsets have been reported. Thesestudies have strengthened the hope that not only interictal states between seizuresbut also specific preictal states exist preceding seizures. The existence of preictalperiods is the basic requirement for genuine seizure prediction