Discriminant Analysis In Depression Based On FMRI Data

Functional magnetic resonance imaging (fMRI) is a fairly new technique that has the potential to characterize and classify brain disorders. It has the possibility of playing a crucial role in designing objective prognostic/diagnostic tools, but also presents numerous challenges to data analysis and interpretation. Based on rest fMRI data about39patients with Major Depressive Disorder and37healthy control subjects, this dissertation conducts a discriminate analysis.It is well known that functional connectivity is an important task in functional imaging studies of the brain. In this paper, significant links are used as features to discriminate health subjects from depression. Feature selection is the key to4005high dimensions feature space consist of functional connectivity. We propose two new feature subset selection methods, which can improve classification accuracy in comparison with the case simply selecting significant features, one is expanding t-test method named Leave-One-Out t-test (LOOTT) feature selection, which can dramatically reduce number of features. while using Support Vector Machine classified method, it achieves80%accuracy by using less than12feature in average. Another one based on Probability Density Function of subjects, can reach a highest accuracy84.21%by using the same classified method. We also show the result of decision tree method applied to discrimination, and find LOOTT can increase the accuracy of tree.A weighted network will be constructed, which consist of links and their weight, use to evaluate contribution of each feature during the classified process, we find four main contributive modules:ORBinf, SMG, IPL--PCG, and the last one is made of some single links. Those functional clusters show important difference between depression and normal group in all, not an single brain area or link, that affords a new horizon for research of functional connectivity.