Impact of Methodological Variables on Functional Connectivity MRI Findings for Autism Spectrum Disorders

Background: Growing evidence suggests that ASD is not a localized brain disorder, but a disorder involving multiple functional networks. Neuroimaging studies of ASD have increasingly focused on connectivity. A large number of functional connectivity (fcMRI) studies have reported network underconnectivity in individuals with ASD. However, there are notable inconsistencies in empirical findings, with some studies reporting overconnectivity in ASD. While some of these inconsistencies can be attributed to the heterogeneity of the disorder, methodological factors may also play a crucial role in differential fcMRI outcomes.

Objectives: To examine how fcMRI results in three ASD data sets may be impacted by methodological variables : temporal filtering, removal of task-related effects, potential group bias in the selection of seeds and regions of interest (ROIs), and whole brain vs. restriction to ROIs in field of view [FOV].

Methods: Three different data sets were used for this comparison: two task-related fMRI data sets for visual search and attention, and one resting-state data set (RS). All MRI data were acquired on a 3T GE scanner with 8-channel head coil. Participants were adolescents with ASD, and matched typically developing (TD) adolescents. All preprocessing and analyses were performed using AFNI. For the task-related data, two pipelines were tested. The first pipeline included high-pass filtering (>.008), whole brain FOV, as well as network of interest FOV. The second pipeline included the following variables band-pass filtering (.008<f<.08), task-regression using a general linear model, and whole brain FOV. Additionally for both pipelines, selection of seeds was based on activation findings for (i.) TD group only, (ii.) ASD group only, and (iii.) both groups combined. For the RS data, the first pipeline was the same as for task-related data. The second pipeline only consisted of band-pass filtering and whole brain field of view. Selection of seeds for RS data was based on the default mode network reported in the literature for TD individuals.

Results: Findings suggested that high-pass filtering yielded stronger group differences than low-pass filtering. Also, task-related pipelines yielded stronger group differences compared to task-regressed ones. With regards to seed selection, seeds based on TD activation (attention task) or on TD literature (RS dataset) were associated with underconnectivity findings in the ASD group, whereas such findings were diminished when seeds were derived from ASD group activation or from activation for both groups combined. For the visual search task, there was evidence in the opposite direction, with the ASD group showing greater connectivity compared to TD group. Finally, FCMRI findings were also affected by FOV. Specifically, greater group differences were found in whole brain versus the network of interest FOV, for all three data sets.

Conclusions: Findings suggest methodological variables substantially affect group differences detected in fcMRI analyses. While there is no clear assumption about the superiority of one approach over another, full disclosure their implications appears crucial in functional connectivity studies when inferences about ‘underconnectivity’ or ‘overconnectivity’ in ASD are made. Combining different methodological approaches may be commendable in future studies for a more comprehensive understanding of connectivity in ASD.