Application Of Structural Equation Modeling To Task FMRI Data In The ABCD Study - Info and Reading Options

The Internet Archive:

A major component of the Adolescent Brain Cognitive Development (ABCD) study is fMRI data from cognitive tasks performed in the scanner, vital to the assessment of neurocognitive development (Volkow et al., 2018). However, the ability of ROI analyses, using general linear models, to extract robust, meaningful relationships between individual differences in task fMRI data and outcome dimensions of interest has been called into question (e.g., Cooper et al., 2018). To address this issue, we will test the application of structural equation modeling to fMRI data taken from activation during the emotional N-back task (EN-back task) to assess whether extracting underlying factor(s) across a prominent brain network underlying cognitive control, the frontoparietal network (FPN), may provide a viable approach to analyzing individual differences in brain-behavior relationships. More specifically, a factor(s) constructed from FPN ROIs should index a latent measure of neurological activation associated with common executive functions (EF); if so, it should predict a latent behavioral factor tapping common EF (see Friedman & Miyake, 2017). This is because areas of the brain that are consistently activated in tasks of cognitive control/executive function, (as in the case of the FPN), may work together to perform general functions such as goal-related processing, including maintaining and adjusting goals, and biasing processing to adhere to goals (Herd et al., 2014). We will test this hypothesis by regressing a common EF factor on our FPN factor(s) via a path coefficient. First, a confirmatory factor analysis (CFA) will be constructed to tap a common factor underlying the brain activation of the frontoparietal network (FPN) when individuals must exert cognitive control, as assessed by the 2-back condition of the EN-back task, as compared to a condition with little or no control demands, as in the 0-back condition. Thus, we will utilize twelve Destrieux parcellations of the FPN (Destrieux, 2010) within the 2-vs 0-back contrast. For model comparison, we will construct a single-factor model and a two-factor model, with two separate factors for the left and right hemisphere. Once we have chosen a best-fit model between these two options, our main analysis of interest is to assess whether the underlying factor associated with the FPN demonstrates significant effect sizes in predicting individual differences in our cognitive-behavioral variables of interest – specifically, latent factors derived from the cognitive battery and task-fMRI behavioral data, including a common EF factor, an updating-specific factor, and an intelligence (IQ) factor. Additionally, we will perform further analyses to explore the robustness and specificity of any observed associations. The first two analyses are designed to determine whether the observed effects of the FPN activation factor in predicting cognitive dimensions are specific to the FPN. First, we will perform permutation tests to determine the extent to which the factor derived from our FPN model better predicts individual differences in our cognitive-behavioral latent factors than a factor representing brain activation across a random selection of brain regions. For example, there may be commonalities in whole-brain activation patterns across random brain regions that are not specifically associated with cognitive control but may still significantly predict individual differences in cognitive dimensions in our dataset. Therefore, the permutation tests allow us to test whether our FPN model performs beyond these commonalities in whole-brain activation. This method will involve performing 1,000 permutation tests of twelve random indicators plugged into a model that replicates the specifications of our chosen FPN model. We are interested in comparing the effect sizes of the factors derived from random indicators in predicting individual differences in our cognitive-behavioral latent factors. Permuted models may not converge due to poor fit – therefore, the final distribution will be from only those models that converged, with the number of models that failed to converge reported. Second, because the regions in this first analysis will be randomly selected and unlikely to index any coherent brain activity, we will also examine the specificity of any observed relationships between the FPN and outcome variables, as compared to a “control” brain network. More specifically, we will extract a factor, using a separate CFA, of activation of the sensorimotor network (SMN) in the same 2-vs 0-back contrast utilizing twelve parcellations of the Destrieux map. For model comparison, we will again construct a single-factor model and a two-factor model, with two separate factors for the left and right hemisphere. From the chosen model, we will determine whether individual differences in activation of this alternative SMN network meaningfully predict our cognitive-behavioral latent factors, as well as whether this alternative network is outperformed by our FPN model. Third, we are interested to what extent the findings with the FPN are specific to our cognitive dimensions by testing a variable from outside the domain of cognitive control. We will do so by testing whether the FPN factor additionally predicts a latent factor for processing speed, using four indicators (Pattern Comparison scores, average go RT for the Stop Signal task, average 0-back RT on the EN-back, and neutral trial average RT on the Monetary Incentive Delay task). Fourth and finally, we are interested in testing whether our proposed analysis approach is feasible for variables outside of cognitive control. To do so we will create a separate CFA to tap the SMN utilizing the 0-back-vs-fixation contrast of the EN-back task, to index change in activation associated with performing a cognitively undemanding task versus staring at a fixation cross, utilizing twelve parcellations from the Destrieux map. For model comparison, we will again construct a single-factor model and a two-factor model. The factor from the chosen SMN model will be tested in predicting the same latent factor for processing speed we have described above, to ensure that the SMN demonstrates meaningful relationships when utilized in a theoretically grounded framework. Additionally, we will test whether the SMN significantly predicts our cognitive dimensions of interest. This analytic approach to examine brain-behavior relationships will be relatively novel, given that fMRI studies rarely have a large enough sample size to power such an analysis (e.g., Elliot et al., 2020). However, a previous analysis from the Human Connectome Project (HCP), which has a sample size of 1,200, has demonstrated the viability of this analytic approach of deriving an underlying factor score to assess brain activation and additionally to link it to individual differences in behavior (Cooper et al., 2018). Therefore, the present work may offer unique evidence in a separate dataset for a novel methodological approach in brain-behavior relationships, as well as possible insight into the nature of such relationships.

Search for “Application Of Structural Equation Modeling To Task FMRI Data In The ABCD Study” downloads:

Visit our Downloads Search page to see if downloads are available.

Find “Application Of Structural Equation Modeling To Task FMRI Data In The ABCD Study” in Libraries Near You:

Read or borrow “Application Of Structural Equation Modeling To Task FMRI Data In The ABCD Study” from your local library.

• The WorldCat Libraries Catalog: Find a copy of “Application Of Structural Equation Modeling To Task FMRI Data In The ABCD Study” at a library near you.