c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 1 3 ( 2 0 1 4 ) 323–337 jo ur nal ho me p ag e: www.intl.elsevierhealt h.com/journals/cmpb A model-based method for computation of correlation dimension, Lyapunov exponents and synchronization from depth-EEG signals F. Shayegh a,* , S. Sadri a , R. Amirfattahi a , K. Ansari-Asl b a Digital signal Processing Lab, Department of Electrical and Computer Engineering, Isfahan University of Technology, 84156-83111 Isfahan, Iran b Electrical Department, Faculty of Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran a r t i c l e i n f o Article history: Received 9 April 2013 Received in revised form 14 August 2013 Accepted 28 August 2013 Keywords: Depth-EEG generator Synchronization Largest Lyapunov exponent Correlation dimension Seizure Accurate feature extraction a b s t r a c t In order to predict epileptic seizures many precursory features, extracted from the EEG sig- nals, have been introduced. Before checking out the performance of features in detection of pre-seizure state, it is required to see whether these features are accurately extracted. Evaluation of feature estimation methods has been less considered, mainly due to the lack of a ground truth for the real EEG signals’ features. In this paper, some simulated long- term depth-EEG signals, with known state spaces, are generated via a realistic neural mass model with physiological parameters. Thanks to the known ground truth of these synthetic signals, they are suitable for evaluating different algorithms used to extract the features. It is shown that conventional methods of estimating correlation dimension, the largest Lyapunov exponent, and phase coherence have non-negligible errors. Then, a parameter identification-based method is introduced for estimating the features, which leads to bet- ter estimation results for synthetic signals. It is shown that the neural mass model is able to reproduce real depth-EEG signals accurately; thus, assuming this model underlying real depth-EEG signals, can improve the accuracy of features’ estimation. © 2013 Elsevier Ireland Ltd. All rights reserved. 1. Introduction Epileptic patients constitute about 1% of the world’s popula- tion [10]. Life quality of epileptic patients depends on the use of anti-epileptic drugs. Long-term usage of these drugs may create unavoidable side effects. Since seizures occur without any obvious herald, the possibility of predicting seizures will open new horizons of remedies to prevent, or at least con- trol, seizures. Reduction of drug dosage, confining the use of drugs to emergencies, stimulation of the vague nerve, and ∗ Corresponding author. Tel.: +98 9133173220. E-mail addresses: farzaneh.shayegh@gmail.com, f.shayeghboroojeni@ec.iut.ac.ir (F. Shayegh), sadri@cc.iut.ac.ir (S. Sadri), fattahi@cc.iut.ac.ir (R. Amirfattahi), karim.ansari@scu.ac.ir (K. Ansari-Asl). other techniques will improve the quality of patients’ lives by reducing their intense feelings of helplessness. Different algorithms have been used to assess whether seizures can be anticipated or not. These algorithms involve extracting some characteristic features from signals, consist- ing of both uni-variate features, like correlation dimension [9,24] and Lyapunov exponents [15], and bi-variate features, like synchronization [32] and dynamical entrainment [16–18]. Three main features that have dominated the seizure pre- diction field during the last years are correlation dimension, largest Lyapunov exponent, and synchronization. 0169-2607/$ – see front matter © 2013 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.cmpb.2013.08.014