Hello Alfonso and everyone,
 
I was recently looking online at some PowerPoint slides that Dr. Whitfield-Gabrieli had written.  In the slides. it was suggested that one could select two sources simultaneously from the Sources column to "aggregate or compare the connectivity results across several ROIs (e.g. to compare the connectivity between LLP & RLP select both sources and enter [1,-1] in the 'between-sources contrast' field)".  (p. 47 of attached file)
 
If I wanted to aggregate right and left BA 46, for example, I would select both, but I'm not sure what the weights would be.  Would they be be .5 and .5?
 
Thanks for any help, (and thanks for all of the great help you have already provided for this great program!)
Mary

Hi Mary,

Yes, that is exactly correct, and the exact contrast values depend on what you exactly intent by 'aggregating'. 

If by 'aggregating' you mean that you want to look at the connectivity with either left- or right- BA46, then you may select both ROIs in the 'sources' list, and enter the 'between-sources' contrast [1 0;0 1] (or eye(2)). The resulting analysis will be a F-test highlighting those regions that show significant association with either left-BA46 or with right-BA46. 

If, on the other hand, by 'aggregating' you want to look at the average connectivity with left- and right- BA46, then you may select both ROIs in the 'sources' list, and enter the 'between-sources' contrast [.5 .5]. The resulting analysis will be a T-test highlighting those regions that show significant association with left/right BA46 (averaging the connectivity values across left-BA46 and right-BA46). 

Independent of the above, you may select any arbitrary between-subjects and/or between-conditions contrasts for more complex comparisons. The interpretation of those will be the same as above (a contrast [1 0;0 1] across sources is equivalent to an OR conjunction of the original analyses across the two ROIs; while a contrast [.5 .5] across sources is equivalent to performing the original analyses on the average connectivity with the two ROIs)

Hope this clarifies
Best
Alfonso



Originally posted by Mary Newsome:

Hi Alfonso,
 
Thanks so much, as usual!
 
May I clarify something? You mentioned selecting between-subjects contrasts while looking at between sources contrasts.  We have two groups in our analysis.  When performing the [1 0; 0 1] between-sources contrast, the same results occur if we have [1 -1] or [-1 1] specified for groups, which made me think the analysis was perhaps collapse collapsing across group (but when All was selected, a different result was obtained).
 
Could you say what the best way of looking at between groups differences when taking an OR approach? 
 
Thanks and best,
Mary 
Originally posted by Alfonso Nieto-Castanon:

Hi Mary,

Your analyses are perfectly correct. If you select two source ROIs (contrast eye(2)) and your two subject groups (contrast 1 -1), the results will show you those regions that show between-group differences in connectivity with any of the two source ROIs. This will be an F-test so the results will be two-sided by nature (you will notice that the explorer shows you an F-stats threshold and the directionality menu is locked to 'two-sided'), so the results will be identical whether you choose a 1 -1 or a -1 1 contrast across groups.

If, after performing these analyses,  you want to know the directionality of the found effects (connectivity differences across groups between the significant clusters and each of your source ROIs), you may do so using the 'explore clusters' button. These are typically reported as post-hoc analyses, and the results may be only significant for one seed ROI or for both, and may show positive or negative effects independently with each seed ROI (the effect sizes displayed here do depend on the sign of your between-subject contrast so if you used a 1 -1 contrast then positive values here will correspond to higher connectivity values in the first group compared to the second).

Another typically useful set of post-hoc analyses involve first importing the connectivity values with each seed ROI using the 'import values' button in your current results explorer, and then using 'tools-> calculator' to further explore the resulting measures (for example, if you find a significant positive between-group difference in connectivity with one seed, you know that that the connectivity values are greater in the first group, but you might still want to know whether that means increased positive connectivity or decreased negative (anti-) correlations in the first group compared to the second; importing the actual connectivity values and displaying them using the calculator allows you to answer these questions and better interpret your results).

Let me know if this  clarifies (or if you would prefer to restrict you analyses to only those regions that show positive or negative between-group differences with any of your seed ROIs -but only if they show the same sign with each seed ROI-; this is a bit more convoluted to test but perfectly possible)

Best
Alfonso
Originally posted by Mary Newsome:

Alfonso, thanks so much.  I noticed I don't see an "explore cluster" button anywhere.  Do I need to upgrade my version?  I'm using 13.l.  (And if I do need to upgrade, would the current analyses be compatible?)
 
Thanks again for everything.
Best,
Mary

Hi Mary,

The "explore cluster" option in version 13.l is still there, only a bit more cumbersome to get to. Simply press the "Export mask" button, and then, when prompted, select the "Explore mask results" option. Unfortunately the "import values" option is a new addition so it was not available in the older version.

If you prefer to update to the latest version that should work without a problem (simply load the old project on the new version and you will have access to these newer options in the results explorer windows). The only option that I believe could possibly require you to re-run your second-level analyses will be if you attempt to use the "export values" option on an analysis that does not include all of your experiment subjects (e.g. a second-level analysis that only looks at one subgroup of subjects). In general the only cases where I would advice against upgrading to the latest version is if you are in the middle of your setup/preprocessing/first-level analyses, just to make sure that you apply the same analysis options to all of your subjects. Other than that, the later versions should always be compatible with projects processed through earlier versions. 

Hope this helps
Alfonso

Originally posted by Mary Newsome:

Hi Alfonso,
 
Thanks very much.  This worked, and as I was exploring the clusters, I was prompted to enter contrast weights in a contrast manager, which had 4 columns.  I had two groups and two sources selected.  Would the first two columns refer to the two groups, and the last two columns refer to the two sources?  And if I wanted to compare the patient group to the control group across the two sources, I would enter 1 -1 0 0?
 
Thanks again,
Mary

Hi Mary,

The appropriate contrast (the one that you entered in the CONN gui to generate the corresponding second-level results) should always appear in that list named as 'connectivity results'. The reason you are not seeing it there is likely because in your case this is an F-contrast, so you need to click on 'F-contrasts' when the SPM contrast manager shows up and you should find it there. 

Just for reference, the SPM design matrix generated by CONN for any arbitrary second-level analysis always has [N1* N2 * N3] total number of columns, where N1 is the number of effects selected in the between-subjects list, N2 is the number of rows in the between-sources contrast defined, and N3 is the number of rows in the between-conditions contrast defined. The interpretation of the different columns in the design matrix follows the same order as above (i.e. the first N1 columns correspond to the subject effects modeled for the first between-conditions contrast and first between-sources contrast, the next N1 columns correspond to the subject effects modeled for the second between-conditions contrast and first between-sources contrast, etc.) In your case, the first two columns correspond to the within-group effects for the first seed/ROI, and the last two columns correspond to the within-group effects for the second seed/ROI, so the appropriate F-contrast to look at between-group differences across any of the two sources should look like: [1 -1 0 0;  0 0 1 -1].

Hope this helps
Alfonso

Originally posted by Mary Newsome:

Alfonso, thanks very much.  I see we are on message #7 now; I sincerely hope you do not get sick of me!

I think part of my problem was that I was thinking this step (the explore mask step) would be like post-hoc t-tests for an omnibus F test, and I would be able to see how the cells differed.  But (please correct me where I am wrong), the results from this step depict the results within each cell, e.g., results for Gp1_LeftBA46, Gp2_LeftBA46, Gp1_RightBA46, and Gp2_RightBA46.  I attached an image of the gui that I think is showing the results within each cell. If these results reveal significant clusters within each cell, could you tell me how I could see where the groups differed?

I'm sorry if I am being thick-headed--thanks for any help!

Mary
Originally posted by Alfonso Nieto-Castanon:

I think I understand now, Alfonso.  I had the between-subjects contrasts with the incorrect weights, which confused things.  Sorry for the extra email and confusion.

Best, and thanks again for all your great support!
Mary
Originally posted by Mary Newsome:

No problem at all, and happy to hear it was just a mis-specified contrast matrix (I imagine it was set to eye(2) instead of [1 -1] so the plots were showing you four values within each cluster -the within-group effects for each of the two sources-, instead of the expected two values within each cluster -the between-group difference for each of the two sources?). 

Best
Alfonso



Originally posted by Mary Newsome:

Yes, you hit the nail on the head exactly, Alfonso.  Thanks so much for sharing your expertise and being so collegial about it.
Best,
Mary 

Hi,

Thank you for this useful post!
I have two related questions:

1) I am using the ROIs from the default atlas file (FSL Harvard-Oxford). Given that the ROIs are split in left and right, if I for instance aim to test bilateral amygdala-hippocampus connectivity, how shall I proceed? If amygdala is my source, as far I as understood, I can select both left and right amygdala and set the contrast to (0.5 0.5]. But what about my target ROI, how do I specifiy that I want both left and right hippocampus?

2) Is it feasible to do network connectivity analyses between salience network and default mode network by combining respective regions into 1? So my seed regions would be all salience network regions, and target would be all DMN regions. For the target I guess I could specify my contrast accordingly, but again I have troubles understanding how I would specify that I want to have a "combined" target region. 

Some help would be immensely appreciated!
Thanks in advance!
Cheers,
Laila