Study On Detecting Intracranial Pressure Based On Feature Fusion Of Resting State EEG Signals

Increased Intracranial Pressure（ICP）is a common clinicopathological syndrome of neurological diseases,which is the common representation of craniocerebral injury,brain tumor,cerebral hemorrhage,hydrocephalus and Intracranial inflammation.The continuous increase of intracranial pressure will cause cerebral hernia and even cause the death of the patient due to respiratory and circulatory failure.Therefore,the timely diagnosis and correct treatment of intracranial pressure increase are very important.Spinal puncture and lateral ventricle puncture are the main methods for clinical measurement of intracranial pressure.These two invasive methods are very painful for patients and have the risk of bleeding and postoperative infection.A non-invasive,accurate and reliable ICP testing system is urgently needed.However,no noninvasive ICP testing system has yet met the clinical requirements of accuracy and stability.Electroencephalogram（EEG）signal contains rich physiological information and pathological information.The change of intracranial pressure will change some characteristics of EEG signal.The purpose of this study is to explore the feasibility of using EEG signal to detect intracranial pressure.The microstate features,nonlinear features,relative power spectrum features and brain network features of resting state EEG signals were extracted from the perspectives of time domain,frequency domain and brain network.The validity of these EEG features for intracranial pressure detection was tested by using support vector machine classifier and regression algorithm.The main work and achievements are as follows.（1）Twenty-three neurology patients,including 10 patients with high intracranial pressure and 13 patients with normal intracranial pressure,were recruited,and their resting state EEG signals were collected and preprocessed.（2）Microstate analysis was used to explore the spatiotemporal characteristics of EEG signals in patients with normal and high intracranial pressure.Results show that the normal patients and patients with high cranial pressure cranial pressure microstate parameter,cranial pressure in patients with A high state of occurrences,duration and time coverage per second than patients with normal cranial pressure,and the state of the D occurrences and time coverage per second were higher than patients with normal cranial pressure,the parameters of A state,D was associated with A significant value of ICP.The above 5 microstate parameters of the EEG signals were used as features to classify patients with high and low intracranial pressure using support vector machine classifier（SVM）.The results showed that all these 5 parameters could achieve effective classification of patients with high and low intracranial pressure,with the highest classification rate of 87%.（3）The sample entropy and approximate entropy of the resting state EEG signals of 23 patients were calculated and statistically analyzed.The results showed that the sample entropy and approximate entropy of Theta band in T3 channel were significantly different between normal and high intracranial pressure patients,and were significantly correlated with ICP values.Using the sample entropy and approximate entropy of the EEG signals in the theta band of T3 channel as the characteristics,SVM can effectively classify the patients with high and low intracranial pressure,and the classification accuracy is 78.3%.（4）The EEG signals of 23 patients were analyzed by relative power spectrum analysis and brain network analysis from the frequency domain and network perspectives.Results show that increasing pressure in the brain of telecom group and normal group of patients with intracranial pressure,theta frequency and beta 1frequency spectrum were significant difference of relative power,and the three parameters with ICP values are significant correlation,the theta frequency band power and beta,alpha 1 frequency band was used as the characteristic,SVM can be achieved in patients with high cranial pressure effective classification,classification accuracy of73.9%.The global efficiency and clustering coefficient of COH network and PLV network in beta band were significantly different between ICP group and ICP group,and there was a significant correlation between ICP and COH network.Taking the global efficiency and clustering coefficient of the COH network and PLV network as the characteristics,SVM can realize the effective classification of patients with high and low intracranial hypertension,and the highest classification accuracy can reach87.0%.（5）The microstate feature,the nonlinear feature,the relative power spectrum,brain network feature of matrix,use the principal component analysis（PCA）for feature fusion,using support vector regression（SVR）to build regression model,forecast intracranial pressure value,the result,all patients with the intracranial pressure value prediction error average 4.47mm H₂O,root mean square error of the mean is 0.65,meet the standards of The Association for the Advancement of Medical Instrumentation（AAMI）.