The Application Of Pattern Recognition Methods In Functional Magnetic Resonance Data Processing

Currently, the life science research focused on the study of brain, while the functional magnetic resonance imaging (fMRI) technique is widely applied in brain research. As the young and fast growing scientific field, fMRI imaging cross-connects with other disciplines closely and covers many aspects of other disciplines. Around the development of two methods in pattern recognition, we explored the methods'application in fMRI data processing in this paper. The main work includes the application of the support vector machine (SVM) and confidence interval in the Multi-Voxel Pattern Analysis (MVPA) feature selection, the design and application of Affinity Propagation Cluster and the self-adaptive Affinity Propagation Cluster.The main contents are as follows:We described the procedure of MVPA, introducing the feature selection methods in MVPA in details, and proposing an algorithm in feature selection method using SVM linked with confidence interval together. We applied the algorithm to the fMRI data and demixed the response fields of left hand motion with that of right hand motion.According to the principle and procedure of Affinity Propagation Cluster, we proposed Affinity Propagation Cluster and took it into the application of simulation and real fMRI data. It solved the problem that the traditional affine clustering method can't handle the large quantity of data, and provided a new unsupervised clustering method in fMRI data processing.According to the Affinity Propagation Cluster as well as the self-adaption processing was used in hierarchical clustering method to produce the self-adaptive Affinity Propagation Cluster and was applied in simulation and real fMRI data. The algorithm solved the problems of invality for affine clustering method handling large quantity of data and the parameter selection bias difficulty for Affinity Propagation Cluster method. As to the personal experience dependent problem, the algorithm managed to objectively select the optimal clustering results.