Independent Component Analysis And Its Application In FMRI

Functional magnetic resonance imaging is a new noninvasive technique to investigation of the functionality of the human brain, which is a powerful tool to study the process of cognizing activity in the brain. This dissertation is devoted to methods of analyzing the fMRI data.1. For the popular software for analyzing fMRI data-SPM, we discuss the mathematical principles of SPM and one of its applications to the research of the functional neuroimage. The methods of estimating parameters and statistical hypothesis testing in general linear model are described. Then we introduce the applications of GLM to SPM and extned the GLM when analyzing fMRI data; At last, the extended GLM is applied to real fMRI experiment data and proved it to be effective.2. Independent component analysis is a new method to deal with fMRI data. We apply a new ICA algorithm-ExtBS and two common ICA algorithms (Infomax and Fixed-Point) to fMRI data, and compare the consistently task-related signals from the aspect of temporal accuracy; the results show that the new ICA algorithom is the best for fMRI data. For the other new ICA algorithm — Orth-Infomax algorithm, we apply it and the two common ICA algorithms to fMRI data, compare the results from temparal accuracy and goodness-of-fit, which shows that the Orth-Infomax aalgorithm can deal with fMRI data effectively and is better that the former two algorithoms.3. The principle of receiver operating characteristic (ROC) analysis and the methods of protracting the ROC curve are introduced. Due to the tradition ROC can only compare the methods for simulate fMRI data, we modify the ROC mothed aimed to ICA algorithoms for real fMRI data; and use the modified ROC mothed to evaluate three ICA algorithoms (FastICA, Infomax and Orth-Infomax).