Research On Impedance Analysis Of Cerebral Blood Flow Based On Deep Learning

Nerve function activity is the basis of brain science research.The process of brain nerve activity is accompanied by the change of local cerebral blood flow.According to the fluidity and electrical characteristics of blood,the change of blood flow will cause the change of bioelectrical impedance in the local brain region,so this change of electrical impedance can be used to reflect the activity of brain nerve.In this paper,a method of characterizing neural activity with cerebral blood flow impedance signal is proposed,and a deep learning model is established to realize automatic classification of cerebral blood flow impedance signal in different physiological states of human brain.First of all,according to the mechanism of brain impedance physiological signal,the experiment of brain function activity induction was designed.Through the experiment,we measured the brain blood flow impedance signal of the subjects in two physiological states,completed the experimental data collection in different brain function states,and constructed the database required by the classification model.According to the characteristics of cerebral blood flow signal,FIR filter,EMD-WT,and coherence averaging technology are used to complete signal preprocessing.Subsequently,according to the original signal and derivative differential signal,the threshold detection method is used to realize the signal waveform extreme point location,signal automatic cycle segmentation and characteristic parameter calculation and analysis.The signal analysis results prove the feasibility of using bioelectrical impedance to detect the cerebral nerve activity.Secondly,a brain impedance recognition model based on SVM and 1D-CNN is proposed.The calculated parameters and segmented signal database are used to realize the automatic classification and evaluation of brain impedance signals in different states.The results show that SVM model and 1D-CNN achieve 82.94% and 85.29% classification accuracy respectively.1D-CNN model omits the steps of artificial feature extraction,and the overall recognition accuracy is higher.However,due to the weak signal response,uneven distribution of sample number and insufficient autonomous feature learning of 1D-CNN,the recognition accuracy of 1D-CNN is not as good as SVM model.On this basis,this paper proposes a signal encoding method based on GAF technology,which maps the one-dimensional cerebral blood flow signal to a two-dimensional representation while retaining the timing information of the one-dimensional signal,expands the characteristic information of the signal,and finally encodes The results are input into the 2D-CNN model for comparison experiments.Experimental results show that the recognition accuracy of the coded recognition model is improved from 85.29% to 92.49%,and the sample sensitivity of the model in the state of functional activity is improved from 75.32% to 87.47%.The model performance is significantly improved,which proves that the proposed method effectively reduces the impact of small signal response on the classification results,and solves the problem of convolutional neural network in one-dimensional time series recognition The model can better realize the automatic classification and evaluation of cerebral blood flow impedance signals under different brain physiological states.