| The ultimate goal for brain functional connectivity study is to propose, test, modify, and compare certain directional brain pathways. Path analysis or structural equation modeling (SEM) is the ideal statistical method for such studies. In this work, we propose a two-stage multivariate autoregressive path analysis/general linear model (MAR-SEM/GLM) approach for the analysis of multi-subject, multivariate time series functional magnetic resonance imaging (fMRI) data with subject level covariates. We also compare this new approach to several existing (and inferior) SEM methods for fMRI data analysis using the visual attention study conducted at the Brookhaven National Laboratory as an example.;In 1890 Roy and Sherrington's paper 'On the regulation of blood supply of the brain' suggested that neural activity was accompanied by a regional increase in cerebral blood flow. Until the advent of fMRI in the 1990's by Ogawa and Lee at the AT&T Bell Laboratories, however, there was no way of non-invasively measuring the flow of blood, and thus the brain functional level, in the cortical areas. Since then fMRI has become an increasingly important tool for the measurement of brain functional activities and brain functional connectivities.;During a fMRI study, each subject's brain functional activity level data are being measured longitudinally and usually more than one brain regions are of interest in each study. Thus for each subject, one obtains a multivariate time series of data from the fMRI experiment. Furthermore, as with most biomedical studies, a group of subjects are usually studied in order to obtain meaningful estimation for the population of interest. Therefore, one would usually acquire a multisubject multivariate times series data set from each fMRI study.;Furthermore, virtually all imaging studies have subject level covariates such as age, gender, education, and measurements of motor, behavioral, and/or cognitive functions using common tests such as the Stroop test for cognitive and Romberg test for motor functions plus various custom-designed self-evaluations forms for behavioral measurements. It is essential to incorporate these "external measurements" or covariates in the path analysis in order to determine the functional relationships between changes in certain brain functional pathways and changes in subjects' cognitive, behavioral and motor functions. (Abstract shortened by UMI.). |
