Multivariate Pattern Analysis Research Of Schizophrenia

Schizophrenia is a common mental illness, which etiology is unclear. The current approach of diagnosis is based on its clinical symptoms, because there is not objective biological markers. Changes in brain structure and function can be found by the traditional unit analysis methods only in group level, which leads to be of limited value to clinical diagnoses. The multivariate pattern analysis(MVPA) that is a better way for clinical diagnoses overcomes the shortcomings through making predictions in an individual level. Our study adopts the multivariable pattern classification as a research method on the function MRI images of schizophrenia patients to explore schizophrenia from two perspectives: the local function and the brain large-scale functional network. We not only establish an individual diagnosis model but also detect the pathophysiology of the disease. In this paper, the main contributions are as follows:1. First we choose the signal of blood oxygen level dependent as our original feature at every local regions. We use the MVPA technique which combines the searchlight algorithm and the principle component analysis(PCA) to select optimal classification features. Then linear Support Vector Machine(SVM) is employed for classifications. The result shows that it has a high classification power in the regions like primary motor cortex, secondary motor cortex, basal ganglia, cerebellum which are associated with motor, anterior cingulated that is associated with emotional and cognition, also include occipital lobe and temporal lobe. The highest accuracy reaches 89.9%. In order to further research the functional change, we also calculate the functional connectivities of these regions. We find that the connections between basal ganglia and many regions which are associated with motor and cognition show significant decrease. More than that, and the correlation of the patient’s connection has a significant negative correlation relationship with duration. The research may provide a new way for the diagnosis and prognosis of schizophrenia.2. In this study we built the whole brain functional connectivity based on Automated Anatomical Labeling(AAL) template. After choosing the connection in the network as classification feature, different feature selection methods: F score, T test, Max-Relevance And Min-Redundancy(mRMR), reliefF algorithms were tested to select features. Experimental result reveals that all the feature selection strategies have achieved a good accuracy, which proves the reliability of the function of the large-scale network. Among them the F score method achieves the highest 88.1% accuracy. The consistent connections and the regions with high weight mainly aggregate in the subcortical nuclei and sensorimotor cortical, which are associated with the cortexstriatum-cerebellum-thalamus loop. This result is similar to the result of the first part, which indicates that no matter from the local or the whole research, it both suggests that basal ganglia has a key role for diagnosis of schizophrenia.This research studies the schizophrenia by MVPA in both local function and global brain network aspects. Both of them achieve the high accuracy and find a significant change in basal ganglia, which means a lot for the research and diagnosis.