تخمين برخط پارامترهاي مدل فيزيولوژيك حمله هاي صرعي از روي سيگنال هاي depth-EEG به منظور استفاده در پيشگويي وقوع حمله

چكيده انگليسي :

Online estimation of the parameters of a seizure physiological model in order to predict seizures Farzaneh Sahyegh Boroojeni f shayeghboroojeni@ec iut ac ir Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan Iran Rasoul Amirfattahi Fattahi@cc iut ac ir Saeed Sadri sadri@cc iut ac ir Karim Ansari Asl Masood Omoomi Abstract Epilepsy has involved about 1 of the world population Abrupt occurrence of seizures makes the epileptic patients lives difficult Motivated by making a better life for these patients many efforts have been made to assess if seizures can be predicted in advance Seizure predictability from one or multi channels of EEG signals both scalp and depth EEG signals have been widely considered but any statistical satisfactory results have not obtained yet Despite the very efforts that have been done until now still false positives of these algorithms are high to be applied in clinic Here we are going to discover if the low accuracy of prediction algorithms is due to the features extracted from signal which weakly reflect the seizure genesis mechanism In this order inhibition and excitation characteristics of neural masses are applied to predict seizures to see whether they can improve the accuracy or not Actually depthEEG signals are supposed to be outputs of a computational model of hippocampus called depth EEG model whose parameters are inhibitory and excitatory characteristics of neural synapses The amounts of neural inhibition and excitation which could not be recorded directly are extracted by using depthEEG parameter estimation algorithm Thus after analysis the depth EEG model and indicating the parameters effective in generating seizure like signals some parameter estimation algorithms are applied to depth EEG signals It is shown that at each value of parameter vector there is a one to one relation between input and output of the model which facilitate the Maximum Likelihood ML estimation algorithm Also this property lets us to define a heuristically parameter estimation algorithm based on synchronization of the model output and the under estimation signal Both algorithms for estimation are time consuming and cannot deliver the results while signals are recorded So to make the proposed seizure prediction algorithm practically useful behavior of inhibition and excitation parameters are modeled by a State Space Model SSM Adding the SSM equations to those of depth EEG model an extended Kalman filter have been used to extract the inhibitory and excitatory parameters in an online manner Eventually these physiological parameters extracted from only one channel are used to discriminate pre ictal and ictal states For depth EEG signals of six patients of FSPEEG database a prediction with 84 04 and 0 28 FPR is obtained According to the surrogate theory these results are statistically significant p value 0 02 to be able to generalized them to vast types of signals Getting these results by using such univariate features inspires the importance of taking seizure mechanism into account for a prediction process Keywords Seizure Prediction seizure generation model model parameter estimation Hidden Markov Model State Space Model Introduction Medically intractable epileptic patients suffer from unforeseen occurrence of seizures In addition to raising the risk of injuries abrupt seizures also make patients feel distress In such situations a reliable prediction of seizures may be the best relief Clinical findings have shown that there is a transient pre ictal phase between interictal and ictal states 1 2 A Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan Iran Electrical Department Faculty of Engineering Shahid Chamran University of Ahvaz Ahvaz IranPDF created with pdfFactory trial version www pdffactory com

شايق بروجني، فرزانه

تخمين برخط پارامترهاي مدل فيزيولوژيك حمله هاي صرعي از روي سيگنال هاي depth-EEG به منظور استفاده در پيشگويي وقوع حمله

پيشگويي حمله صرعي , مدلسازي , مدل مخفي ماركوف , فيلتر كالمن