FMRI-based Brain Pattern Analysis Methods And Applications

Brain is the most complicated information processing system in nature, and brain researches have always been promoting our cognitions of brain and nature. Functional magnetic resonance imaging (fMRI) is one of the most effective techniques on brain researches. The pattern analysis methods of fMRI, making a solid foundation for brain exploration, enable us to acquire available information from brain. This dissertation focuses on the pattern analysis methods of brain fMRI data and their applications in both a depression medication status estimation study and a brain mechanism study on natural scene recognition.The medication status estimation study mainly used multivariate pattern analysis (MVPA) methods. Firstly, the subjects' original resting-state brain functional imaging data were preprocessed by the statistical parametric mapping software package (SPM). Then after temporally filtering and region division, correlation coefficient rank was used between each pair of the region mean time series to get the corresponding connectivity data. Furthermore, the medication status of patients who achieved clinical remission were predicted by the linear support vector machine (SVM) classifier, which was trained with principal component analysis (PCA) on the data of depressed patients and controls. And last, by descending sorting the element of the product of the SPM and the SVM projection matrixs, we measured the contributions of the corresponding features to the classification and reconstructed accordingly the most discriminating functional connectivity networks. Based on resting-state functional connectivity magnetic resonance imaging (rs-fcMRI) data, this study presented to use MVPA methods for depression medication status estimation for the first time, and demonstrated that it can make accurate predictions about medication status of depressed patients with remission. The reconstruction results showed that the most discriminating functional connectivities were located within or across the cerebellum, default-mode network, and affective network, indicating these networks may play important roles in major depression, and suggesting that rs-fcMRI can provide potential information for clinical diagnosis and treatment.The brain mechanism study on scene recognition used SPM and MVPA methods, and the original brain functional imaging data were derived from the fMRI experiment by block design. Firstly, the fMRI experimental task sequence, in which scene images were as stimulus of task, was designed and presented by the design platform of E-prime software. Furthermore, the original signal data were preprocessed by SPM. And last, SPM analysis method, Sparse coding method and Searchlight method were respectively used to estimate brain regions responsible to differentiate complicated scene categories. Based on fMRI data, this study presented Sparse coding method to a new application, the brain functional pattern analysis of natural scene recognition. The experimental results showed that, employing SPM, Sparse coding and Searchlight methods on the subjects' fMRI data, we all achieved finding the areas which contain information that distinguished among six scene categories recognition, and these areas mainly distributed in the primary visual cortex (V1), the parahippocampal place area (PPA), and the retrosplenial cortex (RSC), suggesting that these areas may play an important role in scene apperceive and recognition problem.