Multivariate Analysis Of Pattern Recognition Based On Medical Image Data

Schizophrenia is a kind of the most common mental illness with the main characteristics as discordance between mentation and environment, embodied in change of basic personality and division of thoughts, feelings and behavior. The article makes a comparison between those abnormal functional connectivity in brain from schizophrenia patients and normal subjects, aiming at identifying schizophrenia patients from the normal.The article has a collection of raw fMRI data of sixty-nine schizophrenia patients and sixty-two normal subjects. Normal subjects, as a comparison group, match with schizophrenia patients within age and gender to be required. The paper applies a multivariate pattern recognition analysis to identify schizophrenia patients from the normal and evaluates the reliability of results by permutation test. The basic idea is as follows:Firstly, It constructs Pearson correlation coefficient of the brain network of patient groups and normal groups, and then the most discriminatory features are selected by methods of two sample t-test, risk difference and Kendall rank correlation. After that, It uses support vector machine as classifier with leave-one-out cross-validation to improve the classification performance. Meanwhile permutation test for the rationality of the classification results has accuracy as the statistic. Finally, the paper compares those discriminatory features and brain regions among the results of three different methods.The results shows that the accuracy of classification with two sample t-test for feature selection is83.97%. The accuracy of classification with risk difference for feature selection is83.21%, and the accuracy with Kendall rank correlation is83.97%. It is observed that the efficiency of classification is well. In addition, the efficiency of classification for patient groups is better than that for normal groups by methods of risk difference and Kendall rank correlation for feature selection. At the same time, the paper find that the most discriminatory regions are located in default network, attention network, sensory network and memory network. Those common discriminatory features in the results of risk difference and Kendall rank correlation primarily include a large amount of functional connectivity between these regions such as superior frontal gyrus, cingulate gyrus, angular gyrus, precuneus, frontal lobe, temporal lobe, precentral and postcentral gyrus, supplementary motor area, olfactory cortex, parahippocampal gyrus, amygdala, caudate nucleus,lenticular pallidum, thalamus and so on. What is noteworthy is that functional connectivity in patient groups between precuneus and posterior cingulate cortex is weaker. The conclusion above has important significance to explore neuropathogenesis and help with psychiatric diagnosis.