Study On The Algorithm For The Intraoperative Localization Of Brain Functional Areas Based On Resting ECoG

Intraoperative localization of brain functional areas can guide doctors to remove epilepsy,brain tumor and other lesions to the maximum extent,under the premise of protecting important brain functional areas,which is the key to improve the success rate of neurosurgery.At present,the electric cortical stimulation(ECS)is the most accurate method for intraoperative localization of brain functional areas,however it has a long positioning time,high technical requirements and is invasive.How to locate the brain functional areas accurately,quickly and safely during surgery has been a major problem in neurosurgery all over the world,which needs to be solved urgently.In this paper,the algorithm for intraoperative localization of brain functional areas based on resting ECoG is studied,including acquisition,preprocessing algorithm,feature extraction and classification algorithm of resting ECoG.Firstly,optimize the acquisition method and preprocessing algorithm of resting ECoG.This paper designed a kind of intelligent detector to detect and indicate the three states of contact between cortical electrode and cerebral cortex,including good contact,suspended,and short-circuited electrodes,which could ensure accurate ECoG collection when good contact is detected.Then,this paper used the wavelet decomposition and reconstruction algorithm to filter the resting ECoG,and db3 wavelet was used to decompose the resting ECoG in seven layers,and the signals of d4,d5,d6 and d7 were reconstructed,filtering out the signals containing low-frequency interference below 2Hz and high-frequency noise above 32 Hz,getting the signals with frequency range from 2 to 32 Hz,finishing the filtering of resting ECoG.Secondly,the feature extraction and classification algorithm was designed.At first,db3 wavelet was used to decompose the resting ECoG in seven layers,and the frequency band energy characteristics of four frequency bands(2 to 4Hz,4 to 8Hz,8 to 16 Hz,16 to 32Hz)were calculated.At the same time,6-order autocorrelation model parameters of each segment of preprocessed ECoG were calculated.Then,combine the two characteristics to form a 10-dimensional feature vector as the final sample feature.At last,SVM classification algorithm based on radial basis kernel function and ELM classification algorithm based on sigmoid activation function were designed.Finally,the modeling and testing of the localization algorithm for brain functional regions were finished.Using the preprocessing,feature extraction and classification algorithm to build “The localization algorithm of brain functional areas during surgery based on resting ECoG”.At first,preprocess the resting ECoG of 6 patients and extract feature to obtain feature samples.Then,divided the feature samples into training sets and test sets by 3:1,next,input the training set of each patient into SVM and ELM respectively,and the model was trained by the 10-fold cross validation and analytical method respectively,finally,use each model to test the test set.The experimental results showed that the average classification accuracy of this paper's algorithm was 95.31%,the error detection rate was 4.69% and the average classification time was 5.0150 s when SVM was used as the classifier.When ELM was used as the classifier,the average classification accuracy of this paper's algorithm was 93.75%,the error detection rate was 6.25%,and the average classification time was 0.0135 s.Therefore,using the feature extraction algorithm of wavelet and autocorrelation model,and the classification algorithm of SVM and ELM comparison for the classification of brain functional areas achieved a good result,and using the classification of ELM algorithm can greatly shorten the time and achieve fast,precise and safe results of brain functional areas' recognition.