Localization Algorithms Of Epileptic Focus Based On IEEG

Early diagnosis of neurological disorders is an important research component of the China Brain Project,and epilepsy is of great concern to the community as a common clinical neurological disorder.Epilepsy is a chronic,recurrent transient brain dysfunction syndrome caused by abnormal neuronal discharges in the brain.The high frequency and randomness of seizures leads to suffering and distress,as well as the risk of accidental injury,depression and suicide.Most patients can suppress their seizures with anti-epileptic drugs,but 20-30% of patients eventually develop drug-resistant epilepsy.These patients can only be treated by surgical removal of the epileptic focal area,and accurate differentiation of the epileptic focal area from the normal area during the preoperative assessment is crucial to the success of the surgery.Intracerebral Electro Encephalo Graphy(i EEG)is considered the gold standard for clinical diagnosis of epilepsy due to its high temporal resolution,high signal-to-noise ratio and ability to capture more accurately the rapid dynamics of the brain.In this article,we use the patients’ i EEG signals as a dataset and design algorithms to classify focal and irrelevant areas in three separate directions: the i EEG temporal signal,the directed brain network connections constructed by the channels,and a combination of both,and to give constructive advice to doctors by predicting the category of uncertain propagation areas.The study of i EEG temporal signals is the conventional approach.However,separate time and frequency domain features cannot fully characterize the temporal characteristics of epileptic seizures.In this paper,we adopt wavelet packets decomposition to decompose the original signal and reconstruct five common sub-bands,extract the Power Spectral Density of each band as features,and construct a classification model using the support vector machine of Gauss kernel function.The experimental results show that this algorithm,in addition to having good classification performance and consistency in the prediction of propagation regions,is also able to identify problematic channels within groups through statistical analysis of the classification results for each channel.Temporal signal characteristics alone cannot characterize the direction and impact of causality between brain regions during seizures.This paper uses the Gated Recurrent Unit-based Granger Causality(GRU-GC)algorithm to estimate effective connectivity between patients’ i EEG channels as a way to construct directed brain network connections and extract local features of nodes for channel classification.The experimental results show that this algorithm has better and more stable performance than the first algorithm on all data and is consistent with the first algorithm on the P-group prediction and discovery problem channels.Finally,based on the effectiveness of the first two algorithms,this paper uses the Graph Attention Networks(GAT)model to fuse the temporal signal algorithm with the directed brain network connectivity algorithm.The features of the temporal signals are used as graph node features,and the directed brain network connections obtained based on the effective connectivity are used as the graph structure to construct the classification model.Experimental results validate the effectiveness of the algorithm,which yields the highest classification accuracy on data with a high number of channels.