Method For Computer-Assisted Diagnosis Of Epilepsy By Combining Clinical Data And Neural Computational Models

Epilepsy is a chronic brain dysfunction syndrome,which is one of the most com-mon neurological diseases and characterized by repeated,transient epileptic seizures.Methods for computer-assisted diagnosis of epilepsy refers to extracting useful diag-nostic information from a large amount of incomplete,noisy,and complicated clinical data by modern technologies,which aim to provide references to clinical diagnosis.At present,most of the computer-assisted diagnosis methods are data-driven,whose basic idea is to extract features from epileptic EEGs,and then to combine the classi-fiers to recognize different EEG patterns.However,such data-driven methods are often non-universal because most of them do not model the underlying mechanisms respon-sible for epilepsy.Based on the successful application of neural computational model(NCM)in exploring the underlying mechanisms of various physiological and patho-logical phenomena,this paper focuses on exploring the computer-assisted diagnostic method of epilepsy by combining with clinical data and NCMs.The generation of pe-riodically epileptic discharges,epileptic early seizure detection and epileptic seizure tracking are systematically studied.Details are presented as follows:(1)A novel neural computational model AHE-CM has been built to study the t-riphasic waves(TPW),which is one kind of periodic epileptiform discharges.Firstly,the AHE-CM was constructed based on three modifications of the well-studied Liley model which emulates mechanisms believed central to brain dysfunction in AHE:increased neuronal excitability,impaired synaptic transmission,and enhanced post-synaptic inhibition.Secondly,a model parameter estimation approach was designed,which is based on the frequency histogram of TPW and particle filtering.Thirdly,performances of the proposed method were verified by numerical experiments on 7 AHE patients at Massachusetts General Hospital(MGH).Numerical results show that the proposed AHE-CM not only performs well in reproducing important aspects of AHE-EEG,namely the periodicity of triphasic waves(TPWs),but is also helpful in suggesting mechanisms underlying variation in EEG patterns seen in AHE.In partic-ular,our model helps explain what conditions lead to increased frequency of TPWs.(2)A new D&F model-driven early detection method of epileptic seizure has been proposed.Firstly,the sensitivity measurement of model parameters was defined on the basis of dynamic EEG features and correlation analysis.According to this,a new model parameter selection approach was proposed,which could be applied to select key model parameters.Secondly,the automatic parameter estimation method was designed by combining the grid search method and random walk method.Third-ly,the comprehensive indicators and criteria in early detecting the epileptic seizures were designed according to the selected and optimized parameters.Numerical results on Bonn EEG database and CHB-MIT EEG database demonstrate that our proposed method does a good job in early epileptic seizure detection.The average detection sensitivity,false positive rate and early detection period attain 100%,0.1/h,and 7.1s respectively.(3)A new neural computational model has first been proposed,and then a model-driven epileptic seizure tracking method has been presented.Firstly,the time-delay Wendling model with sub-populations(TD-W-SP)was proposed in three aspects of improvements:(1)a new response function is formulated in pyramidal population;(2)two time delays are added in excitatory interneuron population and slow inhibito-ry interneuron population;and(3)each population is divided into a number of sub-populations.Secondly,combining with two dynamic EEG features and genetic al-gorithm,an automatic parameter estimation method was introduced.Thirdly,the comprehensive indicators and criteria in tracking epileptic seizures were designed according to the obtained parameters.Numerical results on CHB-MIT EEG database demonstrate that our proposed method does a good job in seizure tracking.The av-erage tracking accuracy,pre-estimated duration and post-estimated duration attain 87.25%,9.8s and 2.4s respectively.