Structural Equation Modeling on the Functional Magnetic Resonance Imaging of the Human Brain

This study aims at optimizing Structural Equation Modeling (SEM) analysis in order to accurately estimate the effective connectivity between active brain regions elucidated by the functional magnetic resonance imaging (fMRI) of the human brain during tasks and cognition. Since the empirical fMRI data are of relatively low signal-to-noise ratio (SNR), we are interested in the effects of noises over SEM estimates. First, we use numerical simulation to evaluate the SNR sensitivity of the estimated path coefficients in the SEM. In addition, we also investigate the dependency of path coefficients on the number of data samples, which are practically limited by the relative slow sampling rate (~2 second per volume). Based on the estimated distributions of path coefficients, we quantify the variability of the path coefficients when SNRs and data lengths vary. In this thesis, we also apply the SEM to respective in vivo fMRI experiments to study causal modulations among brain areas during visual cognition and acupuncture stimulus. The SEM analysis developed in this thesis can suggest likely modifications of the a priori directional connectivity required in the traditional SEM and offers statistical inferences on the path coefficients. This method can be used for other brain imaging data.