Functional connectivity of cortex, amygdala, insula – fMRI data analysis using vector autoregression

Abstract

Owing to recent technological advances with high-Tesla MRI scanners, functional imaging of neural tissues with high resolution of the temporal as well as spatial domains comes within reach. Thus, an increasing demand for tools that allow the modeling and evaluation of temporal data, i.e. data that carry sequential information, will likely result. Time series models based on such data can be computed to study the dynamical connectivity of brain structures. We focused on the method of vector autoregression (VAR) by which the strength of sequential interactions among multiple BOLD responses can be assessed, as acquired by fMRI. The method of time series analysis was applied in data sets from 20 subjects listening to auditory stimuli. These stimuli were of an affective nature (a person sobbing; a person laughing) and control stimuli (backward-sobbing, backward-laughter, silence). Each data set consisted of 207 consecutive MR scans. Models composed of 6 variables (i.e., the following regions of interest: Amygdala left/right; Insula left/right; Auditory cortex left/right) were computed. VAR of these variables resulted in a statistically significant model of the sequential interactions among these variables in the sample. It was found that the auditory cortex was directly influenced by the independent variables (the auditory stimuli). Several further interactions were observed, prominently among these an inhibiting effect of the auditory cortex on the amygdala. In addition to these functional results, the methodological merits and limits of the proposed method are discussed.