Study On Detection And Recognition Of Brain Activity Signals In Acute Tonic Pain Based On Microwave Technology

Exploring and uncovering the neural mechanisms of pain brain activity are challenging scientific endeavors.Knowledge on these aspects of brain biology can be used to develop novel medical diagnostic methods and engineering applications.Moreover,it is important for the early diagnosis and prompt treatment of various brain diseases,especially of the neuropathic pain.Some of the methods for monitoring pain brain activity and early detection of brain diseases are electroencephalogram(EEG),magnetoencephalogram,blood oxygen level-dependent functional magnetic resonance imaging,and positron emission tomography.These method can improve the detection and monitoring of various brain activities due to pain to a certain extent.Furthermore,they offer opportunities for developing pain treatment and brain diseases surveillance.However,these methods suffer from a myriad of technical difficulties,such as low temporal or spatial resolution,requirement of expensive equipment,high detection cost,and radioactive damage to human body.Therefore,they cannot be widely used as an early warning system prior to hospital admission,as well as for accurate identification of pain signals.The microwave detection technology has gradually attracted the attention of both local and foreign researchers because it is not limited by temporal and spatial resolutions,it is portable,and it is inexpensive.It is widely used in testing for stroke,brain tumors,and insomnia prior to hospital admission.In this dissertation,we monitored the brain activity arising from acute tension pain via the microwave detection technology.We comprehensively investigated the effects of changes in microwave frequency on the activity of nociceptive neurons,brain activity due to pain,and signal recognition.The main contributions of this dissertation are summarized as follows:1)The effects of microwave radiation on relative EEG power were evaluated using microwave characteristics that can regulate the discharge frequency of brain activity functional sites.On the basis of the relationship between microwave radiation and brain dynamics,a method for measuring the relative power of microwave emission frequency and dynamic brain oscillation rhythm and for monitoring changes in source localization activity was established.The effective microwave detection frequency was then determined.By calculating the changes in the relative power of dynamic EEG frequency band and by observing the effects of s LORETA on the source localization of neural activity,the inhibition-activation relationship between different microwave frequencies and the firing frequency of the neurons in the activated brain functional sites was verified.Results demonstrated that microwave radiation could change the relative EEG power.Moreover,the source activity would either increase or decrease with the change in microwave frequency.Analysis of the results revealed that 5 GHz was the best detection frequency.Furthermore,the waveforms and the spectra of microwave transmission signals were compared with those of pain EEG signals.The correlation coefficient between microwave transmission signal and EEG signal was calculated by linear correlation analysis to verify the feasibility of microwave detection of brain activity due to pain.2)An entropy combined with machine learning method was presented to identify pain information in microwave-transmitted signals.According to the time series variation characteristics of microwave transmission signals,multiscale fluctuation-based dispersion entropy based on time-domain variation and power spectrum Shannon entropy based on frequency-domain variation were proposed as the features of “no pain” and “pain”signal complexity,respectively.The entropy features of “no pain” and “pain” were extracted via empirical mode decomposition and variational modal decomposition.The minimal-redundancy-maximal-relevance criterion based on mutual information was used and principal component analysis was performed for feature selection and dimensionality reduction,respectively.Support vector machine,k-nearest neighbor,linear discriminant analysis,and naive Bayes were employed to train and classify feature datasets.The shallow machine learning model was selected to train and classify the feature dataset.The classification performance of feature extraction algorithm and feature selection algorithm was analyzed.Results showed that entropy could effectively differentiate pain signals with different complexity.Compared with PCA,the feature data obtained via variational modal decomposition combined with the minimal-redundancy-maximal-relevance criterion had the highest classification accuracy,sensitivity,specificity,and positive predictive value.It is of great value to improve the identification performance of microwave detection.3)A new method for recognizing brain activity due to microwave pain perception based on multiple types of composite features was proposed.In the“no pain”and“pain”binary classification task,wavelet packet decomposition and variational modal decomposition were used independently or superimposed to extract relative energy change,refined composite multiscale dispersion entropy,refined composite multiscale fuzzy entropy,and autoregressive model coefficient based on Burg algorithm as the composite features of “no pain” and “pain”.The shallow machine learning classification strategy was adopted to evaluate the performance of the composite features.Results showed that the superimposed feature extraction algorithm could achieve greater recognition ability and could more likely capture the nonlinear dynamics of the brain from the signals than the other algorithms.The random forest classification strategy was better and could obtain more stable results than the superimposed feature extraction algorithm.The proposed method could further optimize the recognition rate of brain activity signals arising from sensation of pain.4)A pain intensity feature representation and recognition method based on deep neural network was proposed.Wavelet packet transform,multiscale entropy,convolutional neural network of different depths,and short and long short-term memory networks of different layer structures were used for feature extraction and classification owing to their excellent ability to subdivide frequency bands.Deep learning and existing shallow learning methods were utilized to design seven feature extraction and recognition models.Results showed that the feature extraction method of convolutional neural network could remarkably lessen the difficulty of distinguishing mild pain from moderate/severe pain.Moreover,this feature extraction method could improve classification performance by over3% compared with the multiscale entropy feature model.In terms of classification performance,convolutional neural network and long short-term memory network,especially bidirectional long short-term memory network,exhibited a higher classification accuracy than shallow machine learning methods.The proposed method resolved the cost of complex feature engineering calculations of pain intensity and enhanced the recognition ability of pain signals of different levels.The microwave detection method proposed herein is not limited by temporal and spatial resolutions.Thus,it can more accurately,promptly,and efficiently detect brain activity due to pain than the other detection methods.It is safe,inexpensive,convenient to use,and can be performed quickly.In addition,this method uses an advanced machine learning technology to mine the information contained in the brain activity data,thereby substantially improving the recognition ability of pain perception and pain intensity.It provides a scientific basis and technical support for the high-precision analysis of brain activity due to pain.