Research On Probabilistic Collaborative Representation Of Epilepsy EEG And Deep Convolutional Characteristics For Seizure Prediction

Epilepsy is a chronic neurological disorder characterized by abnormal and excessive electrical discharges of neurons in the brain.The physical and psychological damage caused by epileptic seizures seriously affect people’s daily lives.The electroencephalogram（EEG）can collect the electrical activities of the brain,which has become a reliable tool for the diagnosis of epilepsy in the clinic.However,the large amounts of EEG data are required to be identified by visual observation of well-trained neurophysiologists.Therefore,the development of automatic detection of EEG recordings provides a new way to alleviate the pressure of neurologists.In addition,seizures can be controlled by antiepileptic drugs or surgery for about 75%of the patients.Unfortunately,the remaining 25%of patients must live with uncontrollable seizures.Therefore,a reliable seizure prediction system can not only improve the quality of their life but also lead to some novel therapeutic methods such as implantable drug delivery or electric stimulation of vagus nerve.In this paper,a kernel version of the robust probabilistic collaborative representation based classifier（R-ProCRC）is proposed for the detection of epileptic EEG signals.The kernel R-ProCRC jointly maximizes the likelihood that a test EEG sample belongs to each of the two classes,and the kernel function method can map the EEG samples into the higher dimensional space to relieve the problem that they are linearly non-separable in the original space.The discrete wavelet transform and differential operation are first employed to process the raw EEG signals.Then the preprocessed test EEG sample is collaboratively represented on the training sets by the kernel R-ProCRC and it is categorized by checking which class has the maximum likelihood,i.e.yields the minimum represented residual.At last,a postprocessing is deployed to reduce misjudgment and acquire more stable results.This method has been evaluated on two different types of EEG databases and shows superior performance.In this paper,a novel method combining local mean decomposition（LMD）and Convolutional Neural Network（CNN）is proposed for seizure prediction.Firstly,the LMD is employed to decompose the raw EEG signals into a set of product functions（PFs）.Subsequently,three PFs（PF₂-PF₄）are selected to learn the EEG features automatically using the deep CNN.In order to obtain the most important information from the features extracted by the CNN,the principal components analysis（PCA）is used to remove the redundant features.After that,these processed features are fed into the Bayesian linear discriminant analysis（BLDA）for classifying the cerebral state as interictal or preictal.Eventually,post-processing is applied to reduce the false prediction alarm.The proposed seizure prediction method was evaluated on the Freiburg dataset and achieved a sensitivity of87.7%with the false prediction rate of 0.25/h using intracranial EEG signals of 21 patients.The experimental results indicate that the proposed method can predict seizures above the chance level and may become a potential approach for predicting the impending seizures in clinic application.The automatic detection and prediction technology of epilepsy proposed in this paper helps to promote the research of the probabilistic collaborative representation of EEG signals and the automatic extraction of feature directions by using deep learning.It also helps to promote the development of relevant theories and the research of detection and prediction algorithms in clinical applications.The algorithm proposed in this paper needs to be tested on more actual EEG data to verify the feasibility of its clinical application.