I have completed seed-based resting state analysis by extracting ROI time series with 3dmaskave and using 3ddeconvolve to obtain the errts dataset which was then fed into 3dttest to look at group differences between young and aged monkeys. These monkeys have been tested on a working memory task and I wanted to look at correlations between connectivity and their behavioral performance. I have seen others use 3dRegAna for this type of analysis but according to recent posts on message board, this is obsolete and 3dttest or 3dMVM is recommended. Can you please advise what would be the best method to look at the relationship between connectivity and cognitive performance?
Thanks Gang!
I just want to make sure I am interpreting the results from 3dttest w/ covariates properly. The sub-bricks that reference means are the differences between groups with the covariate as a regressor (i.e. removed from the analysis so it’s not contributing to the resulting differences)?
The sub-bricks looking at the difference between groups and the covariate (as help function says: difference of slopes wrt covariate IQ - or in my case, cognitive performance - Young-Aged_behavior) means that there are significant differences between the groups that is dependent on the behavioral score?
To visually show this would I have the Young-Aged_behavior sub-brick as the overlay and threshold with the corresponding z-score?
In addition to the fact the two groups had marginal difference in terms of average cognitive performance, are the two groups expected to have some extent of cognitive difference? If so, you may consider performing within-group centering, which may have some impact on the inferences on group differences of intercept (correlation with the seed) in the brain.
To visually show this would I have the Young-Aged_behavior sub-brick as the overlay
and threshold with the corresponding z-score?
Aged animals can generally be broken down into 2 groups - high performers that perform on par with young and low performers who have cognitive deficits - so this is a little tricky. The results look very similar whether I center SAME or DIFF - the only difference is the cluster size for correction is larger (very large) when it is centered as DIFF. Any opinion/suggestions which would be the most appropriate? The results make a lot of sense biologically.
Also - If what these results are saying is that the slopes are different with respect to performance, how do I know about directionality of each of the groups? Is there a way to extract a connectivity score or signal change value that can be plotted against cognitive performance?
Which effect is your focus: behavior effect or the correlation with the seed? Or both? The centering issue is only applicable for the correlation with seed (intercept), not the behavior effect (slope).
Aged animals can generally be broken down into 2 groups - high performers that perform on par
with young and low performers who have cognitive deficits - so this is a little tricky.
If you’re referring to group difference about the correlation with seed, it depends on what kind of questions you want to answer: do you want to compare the aged animals with high performance to the young animals? Or do you want to compare the aged animals with low performance who have cognitive deficits with the young animals? Or, you just want to the whole aged group and with the young at some common behavior value?
If what these results are saying is that the slopes are different with respect to performance,
how do I know about directionality of each of the groups?
Do you see the behavior effect (sub-brick) for each group in the output?
We are interested in the behavioral effect as we have shown that aged monkeys have increased connectivity with various seed regions involved in learning and memory. This analysis was done with 3dttest++ without covariates but I get similar results with covariates if centered as different. We are unable to compare the 2 aged groups due to lack of statistical power so we wanted to see if functional connectivity varies with behavioral performance across all ages. I realize your new bayesian method might work for these type on analyses (and you had run some analyses for us on a different, even less powered group of animals - in hindsight, I should have given you these datasets instead). At this point, however, we are looking for something simple akin to what was published in Ash et al. (https://www.pnas.org/content/113/43/12286). Here, they also extracted some connectivity data to plot against performance but I am not quite sure how they did that. Any ideas there?
In theory, comparing all of the groups you suggested would be ideal, however, I don’t think that is feasible with the number of animals we have. We have done each unimpaired vs. young and impaired vs. young using 3dttest++ without covariates and show different Fc maps when each group is compared to young - I thought this would indicate that significant clusters/regions/activity observed in unimpaired but not impaired might be important for cognition.
Yes - I see a Young_behavior and Aged_behavior. Young_behavior has significant clusters while Aged_behavior does not. Would this indicate then that young animals have increased connectivity between the seed and these positive clusters and that this connectivity is important for cognitive performance? Does this say there is no relationship in aged animals and just young?
Young_behavior has significant clusters while Aged_behavior does not. Would this indicate then
that young animals have increased connectivity between the seed and these positive clusters
and that this connectivity is important for cognitive performance?
At least you have some strong evidence for the positive effect of cognitive performance at those regions for the young group.
Does this say there is no relationship in aged animals and just young?
No, for two reasons: 1) no strong evidence of an effect is not the same as an evidence of no effect, and 2) you’re assuming linearity for the effect of cognitive performance. For 1): if you set a threshold of voxel-wise p-value of 0.05 or even 0.1, do you see something meaningful? If you’re willing to focus on a list of regions, a region-based approach may allow more exploratory analyses (e.g., nonlinear relationship).
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.