Magnetic Resonance Imaging Study On Depressive Patients With Anxiety Disorders

Although depression and anxiety disorders are two independent diseases, they often coexist with each other, which is called "comorbidity” in medicine. Depressive patients with anxiety disorders(referred as "comorbid patients") have many features, such as late age of onset, severe illness, high suicide risk, and badly damaged social function. Based on structural magnetic resonance imaging(MRI) of the human brain, we studied brain structural morphology of comorbid patients by using the methodologies of morphological analysis and machine learning.The main content of this paper consists of two parts:Firstly, we applied voxel-based morphometry(VBM)and region of interest(ROI) analyses to investigate the differences of gray matter volume and white matter volume among three groups, including comorbid patients, depressive patients, and healthy control. The result indicated that significant differences of gray matter volume were found primarily in the insular cortex, frontal lobe, parietal lobe, and temporal lobe. Besides, significant differences of white matter volume were found predominately in the cerebellum parietal, frontal, and parietal lobe. Moreover, most of the brain areas with significant differences among groups showed a significant correlation between clinical evaluation and regional volume of gray matter or white matter. For example, the gray matter volume in the left insular cortex was significantly correlated with Hamilton anxiety score, indicating that the left insular cortex may be an important brain region to distinguish depressive patients from comorbid patients.Secondly, we studied the application of machine learning theory on MRI data. We performed classification training and testing of MRI data by using the standard support vector machine(SVM), SVM with F-score feature selection, and SVM with RFE feature selection. By comparing testing results of three classifiers, we found that the methodology of SVM with RFE feature selection showed best performance. For example, in the classification test of depressive patients from comorbid patients, we used the data combing gray matter volume and white matter volume and applied the method of SVM with RFE feature selection; we achieved the excellent results of that the specificity, sensitivity and accuracy are all up to 100%.In conclusion, compared with depressive patients and healthy control, comorbid patients showed gray matter and white matter abnormalities primarily in the parietal lobe cerebellum, insula, frontal lobe, and parietal lobe. We also found significant correlations between clinical evaluation and gray matter or white matter abnormalities. Machine learning analysis using the data of structural MRI achieved good performance in the classification of depressive patients, comorbid patients, and healthy control.In this thesis, we analyzed structural MRI data of comorbid patients by combining morphological analysis and machine learning, found the specificity in brain morphology of comorbid patients compared with depressive patients and healthy control, and achieved high accuracy in automatic classification among three groups. The findings in the thesis may provide a potential biomarker for disease diagnosis of mental illness.