The Study And Application Of Classification Method Based On Non Parametric Statistics

Brain functional magnetic resonance imaging (fMRI) is a new technology in the development of medical imaging since 1990s. It is usually used by researchers to study the changes of the functional activities of the brain areas. The research and application of fMRI data using pattern recognition method have become one of the important directions of brain science study. The functional connectivity of brain network structure based on AAL template, Which as a distinguishing feature of the patient and the normal person, is an important content in the study of brain functional imaging. But it is a high dimensional data to construct the functional connectivity in the whole brain region, so that there may be some redundant information in the classification process, so the feature selection is the key. In addition, the classification process in the selection of the classifier is also playing a crucial role.The resting state fMRI data of 39 patients with depression and 37 normal controls were analyzed in this study. To contrast of the original data without any examination of the selected feature and the application of the T test. The main idea of this article is different from the previous studies of parametric test pick significant features, the application of two kinds of non parametric tests (the K-S test and the M-W U test) select significant features, the original feature space dimensionality is greatly reduced. Then three classifiers (Fisher discriminant, KNN, naive Bayes) were used to judge the patients and normal persons. The results show that the classification accuracy of the classifier combined with non parametric test is higher than that of the parameter T and the effect was significantly improved. Data of the three kinds of classifier to select different feature number respectively in discrimination, found the highest accuracy rate was higher than that of 76%, Among them, the highest accuracy rate of three classifiers combined with the M-W U test is more than 80%, Fisher discriminant analysis combined with K-S test, the accuracy rate is 81.58% when the number of features is reduced to 10,When the number of features is 46, the accuracy rate is 90.79%.In this paper, we get the optimal discriminant — the Fisher test criterion with the K-S test through the results comparative analysis, and there is proved the reliabilityof this discrimination method through the permutation test. At the same time, we get the most discriminative brain regions through this feature selection method, which mainly include:the right inferior frontal gyrus, the superior frontal gyrus, the parahippocampal gyrus, the dorsolateral, the cingulate gyrus, the fusiform gyrus, the inferior parietal gyrus, the supramarginal gyrus, the caudate nucleus in brain area, the right inferior frontal gyrus among these areas share the maximum weight. These brain regions correspond to the significant difference between the brain areas of patients and normal human, which provides a theoretical basis on pathological study of depression patients, which also has practical significance in assisting doctors of medical clinical diagnosis.