Research On Feature Selection And Classification Method Of FMRI Data Based On Statistical Information

Functional magnetic resonance imaging(f MRI) data analysis has been a hot topic of research in the field of cognitive neuroscience, f MRI data classification as a basic task of f MRI data analysisand has been widely attracts the attention of the scholars of domestic and abroad. However, most of the research efforts adopt the univariate analysis method, and there is almost not intensesive research on the f MRI data, in particular, these existing works ignored the influence of the high dimensional and sequence variation characteristics of f MRI data on identifying results. This paper aimed at these problems, combined with characteristics of f MRI data, based on the main line of feature selection and classification method are carried out from the following two aspects:(1) To solve the problem that high dimensional property of f MRI data, we take into account interactions between multiple features and spatial patterns, a new feature selection method(named as "NMI-F") was proposed by sequentially combining the normalized mutual information(NMI) and fisher discriminant ratio. In NMI-F, algorithm based on spatial correlation between the multiple features of f MRI data, the normalized mutual information was firstly used to evaluate the correlation among features. Secondly, fisher discriminant ratio was then applied to choose this feature as the one that has the power of discriminating between cognitive status to decode variables of interest from f MRI data and thereby show the data contain information about them. Experimental results on two public f MRI datasets show that compared with some other classical feature selection, our algorithm represents preferably performance in several assessment metrics.(2) To solve the classification performance is restricted by the change of f MRI sequence variation, this paper proposes a support vector machine classification method based on f MRI voxel sequence variation and ensemble feature selection. Firstly,in feature extraction, the feature is extracted by using the sequence variation of voxel, and the initial set is obtained by using top-k strategy. Secondly, the initial set are further optimized through ensemble feature selection based on statistics. Finally, the block coordinate descent approach for support vector machine classification to classify. Tasks related dataset experiment show that the new method can decode brain mental states, and has obviously competitive compared with some other methods.This paper achieves two novel methods of feature selection and classification from f MRI data, which provides useful reference for the research of f MRI data classification and help further development of neurocognitive.