Research And Application Of Human Brain Emotional Mechanism Based On Brain Network

Emotion is an important psychophysiological phenomenon of human beings.Bipolar disorder is a serious psychiatric disorder characterized by polarized emotions.Moreover,the suicide rate of people with bipolar disorder is extremely high,which can cause huge losses to individuals and society.Therefore,this thesis takes the brain network of patients with bipolar disorder as an entry point,and combines MRI data and EEG signals to study the emotional mechanism of the human brain.The main work is as follows:（1）This thesis constructs the structural network of bipolar disorder patients and normal people based on gray matter.It was found that the assortativity of the patient network was significantly reduced,and the attributes of nodes such as the middle temporal gyrus,parahippocampal gyrus,cerebellar vermis,amygdala,and insula were abnormal,and it was inferred that these abnormalities were related to the emotional polarization of patients.Simulations were performed when the structured network was attacked,but no differences in network resilience were found between subjects.The network center was studied from two aspects of degree value and betweenness centrality,and it was found that the number of centers increased in patients,and they were mainly concentrated in the cerebellum and middle temporal gyrus,while the amygdala was no longer the center.This suggests that the middle temporal gyrus and cerebellum may play an important role in emotion regulation.（2）Based on the pearson correlation coefficient and partial correlation coefficient,the functional networks of the two groups of subjects were constructed in this thesis,and it was found that there were differences in the assortativity（disassortativity）of the patient network,which was consistent with the research conclusion of the structural network.It showed that the change of homomatch may lead to the decline of emotion regulation ability.At the same time,the ALFF,Re Ho,and VMHC of the brain were analyzed,and it was found that the VMHC values in the precuneus,temporal lobes and other brain regions of the patients decreased significantly.（3）Based on the above information,this thesis trains a machine learning model and confirms that the characteristics of different brain regions can achieve effective classification of patients.Then,reduce the dimensionality of gray matter features based on PCA,filter the remaining features based on mutual information,and integrate SVM,DT,and GBDT based on the voting method and stacking strategy,this thesis gets two ensemble models.The final classification accuracy of voting model reaches 84.5%,which can be initially used for auxiliary diagnosis of bipolar disorder.（4）This thesis selects 10 EEG channels based on abnormal brain regions and conducts emotion classification research on the DEAP dataset.Firstly,emotion classification is performed on 32 channels and 10 channels respectively,which proves the effectiveness of channel selection.After that,this thesis also designs and implements the convolutional bidirectional recurrent attention（CBRA）model based on convolutional neural network,bidirectional recurrent neural network,and attention mechanism.The average accuracy rates in the two dimensions of valence and arousal are 89.2% and 88.2%,respectively.（5）This thesis also extracts the music features of DEAP,implements a music classification model using SVM,and screens music based on this model,which is used as the emotion-inducing material for EEG acquisition experiments.In the CBRA model,the average accuracy rates in the dimensions of valence and arousal are 77% and 76.4%,verifying that music with different characteristics can stimulate the human brain to generate different types of emotional EEG,which can be used for emotional regulation brain-computer interface.