Novel imaging-processing-based analysis of fMRI data

Since its development in the early 1990s, functional MRI has emerged as a useful tool to explore the functional behavior of the human brain. Image processing for fMRI data analysis has been playing a very important role for determining which parts of the brain are activated by different types of stimuli. In my thesis, two novel image-processing-based approaches have been proposed. The first approach adopts three partially spatial temporal adaptive processing (STAP) schemes to reduce dimensionality of the fully STAP algorithm and make it more tractable. Computer simulations incorporating actual MRI noise and human data analysis indicate that these three partially adaptive STAP algorithms, especially element space, can attain the performance approximating that of fully adaptive STAP while significantly decreasing the processing time and maximum memory requirements. The second approach presents the idea of transforming the fMRI activation detection problem into a traditional image segmentation problem. With the incorporation of a Bayesian image segmentation technique, the expectation-maximization/maximization of the posterior marginals (EM/MPM) algorithm, to signal detection for event-related functional MRI (fMRI), the critical prior information in the spatial domain is preserved, overcoming a notable drawback of conventional fMRI analysis. The proposed EM/MPM-based approach is demonstrated to enhance detection performance over traditional GLM and ICA analysis, and yields activation that is comparable to those methods, suggesting that the new analysis procedure is a viable option for future use and development in the context of fMRI analysis. The success of this EM/MPM-based method will enable the very rich collection of advanced image segmentation techniques to be applied to detection of fMRI activation