Research On Pattern Recognition Method Based On Multivariate Brain Network And Its Application In Depression

Major depressive disorder(MDD)is a common mental disease characterized by dysregulated emotion.Exploring the neural activity of functional brain network of patients with MDD in temporal and spatial characteristics is of scientific significance to reveal the abnormal mechanism of functional separation and integration in patients with MDD and to help the diagnosis and evaluation of patients with MDD.However,previous studies mainly focused on the univariate and single level imaging features of brain functional activity in patients with MDD.In this paper,we used multilevel support vector machine model and multilevel community detection algorithm based on resting-state and task-state f MRI data.It reveals the abnormal neural response pattern and dynamic module attributes of patients with MDD from static and dynamic dimensions.Then,we extracted relevant features as neurobiological markers to evaluate the prediction power of the disease severity of patients with MDD.The main contents are as follows:In the first study,we used the generalized linear model and psychophysiological interaction analysis to calculate the whole-brain task-evoked activation and taskmodulated connectivity of patients with MDD and healthy controls with the audio-visual emotional task-f MRI.Then,three different levels of functional features were extracted and the dimensions of the features were reduced by the feature selection framework with filtering-method and embedding-method.Finally,suppot vector machine classification and regression models were used to identify patients with MDD from healthy subjects and to assess the prediction power of the clinical scores of patients with MDD.The results showed that both task-evoked activation and task-modulated connectivity could effectively classify and evaluate patients with MDD,and the multivariate features integrated by activation and connectivity further improved the classification and prediction performance.In addition,the most discriminative features were mainly located in the emotional network,such as prefrontal cortex,limbic network and striatal cortex,indicating that abnormal neural activity patterns appeared in the audio-visual emotional processing of patients with MDD.In the second study,we used the dynamic sliding-window method and Pearson correlation analysis to analyze the dynamic functional connectivity of the nodes in emotional network of patients with MDD and healthy controls with the resting-f MRI and task-f MRI.Then,community detection algorithm was used to reveal the flexibility,recruitment and integration of nodes in multi-layer network and clustering algorithm was used to explore the community module attributes in different states.Finally,random forest model explored the ability of dynamic module characteristics to predict clinical scores of patients with MDD.The results showed that patients with MDD exhibited abnormal dynamic communication and emotional control dysfunction in different emotional states.Compared with healthy subjects,patients with MDD showed higher flexibility but lower recruitment in prefrontal network during the resting state while showed lower flexibility and integration but higher recruitment in limbic system and striatum network during the task state.In addition,both the three dynamic module features could effectively predict the disease severity of patients with MDD,revealing the aberrant dynamic transformation mechanism of emotional network system in patients with MDD.