Research And Application Of Voxel Based FMRI Data Classification

The function of the brain is to receive and process information. As a long time, people are increasingly obsessed with the objective of the brain. With the development of science and technology, various of technology which is used to study the brain are came into being, along with vast amounts of data generated. It can not dig out hidden behind the message with simple process. Therefore, we must use effective methods and advanced machine learning methods to analysis the data. In order to discover new lays,reveal the mysteries of the brain cognitive activity.Functional magnetic resonance imaging (fMRI) is a rapid development in recent years of an imaging technique. It has high resolution, non-invasive advantages for the study. It provides a lot of convenience for the study of brain, especially in the brain area location and brain network.This study is part of the National Natural Science Foundation Project. It is main to explore and research the color images based on image characteristics in the brain activation patterns.Designing rational experiments, select suitable feature selection and feature extraction method for the study, select appropriate classification algorithm fo data, to identity data based on different features. The main work is as follows:Exploring the common mode for psychological experiments and fMRI experimental. Thinking about the purposes of this study, We design the rational paradigm, select the appropriate subjects, and set a reasonable device parameters based on experimental requirements. Studying the data processing methods, reasonable methods are selected to process data from the experiment, data processing,feature selection, feature extraction and classification. SPM software is used which is based on MATLAB platform.In feature selection, comparing feature activation area and single voxel integrated, finding that the single voxel the integrated feature selection method is more suitable for this study.Feature extraction, the comparison of the maximum change in BOLD, BOLD change speed, BOLD and BOLD cumulative value of the time series variance and mean combination of four feature extraction method. In the four methods, the last is more suitable for the present study. In classification algorithm, comparing the SVM with Adaboost, the last is found with a higher accuracy rate close to90%.This study showed that the use of fMRI data can be obtained human visual information, which can effectively predict the feature binding mode and the complexity of the image. The Brodmann19region is more sensitive with the complexity of image, it is more closely to the task. The research method can be applied to BCI (brain computer interface, BCI) and other fields.