Print

Print
Mark,

Please see inline responses below.


On Thu, Dec 12, 2013 at 1:38 PM, Mark <[log in to unmask]> wrote:
Hi,

I've checked serial posts but still get confused so I hope someone could help me. Sorry it's a bit long.

I'm interested in a ROI and we hypothesized an important role of this ROI in our task. In my group-level analysis [17 subjects, with 3 covariates in the model, 1 of which was covariate of interest (we want to know which region's signal covaries with this behavior score) and 2 of which was covariates of no interest (we want to regress out age and body height)], if I used whole-brain FWE p=0.05, this region was not activated. It was not surprised because this subcortical region is tiny.

Based on a previous study, I used MarsBar to create this ROI image file, which is a 10mm-radius sphere centering at (0 -32 -10).

<Q.1> I used 2 methods to do SVC but wonder which one should be correct or better?

Method I:
In group-level analysis, I first output "Results" and used p=0.005 uncorrected. In spm8, I then clicked "small volume" and selected this ROI as image. the results table showed:

cluster-level: Pfwe-corr = 0.016
    peak-level: (1). Pfwe-corr = 0.020, T = 4.36, coordinates at (-2 -28 -8)
                     (2). Pfwe-corr = 0.044, T = 3.78, coordinates at (4 -28 -16)

Method II:
When I ran the group-level analysis, in "factorial design specification," I entered this ROI image as an Explicit Mask. After estimation, I output the results by selecting FWE p=0.05 (rather than 0.005 uncorrected used above). Then the results were slightly different:

cluster-level: Pfwe-corr = 0.015
    peak-level: Pfwe-corr = 0.028, T = 4.36, coordinates at (-2 -28 -8)

>>> The difference is in the computed smoothness of the data. In Method 1, the smoothness is computed from the entire brain mask, in method 2, the smoothness is computed only within the mask. The smoothness estimate determine the size and number of resels, which changes the corrected statistics. Notice that the T-statistics are the same; however, if you have unequal variance, then the t-statistics may also differ slightly due to different variance corrections. 

There isn't a "right" way and a "wrong" way. Method 1 would also allow whole brain exploratory analysis or to change the mask later, which is why I prefer it over Method 2.

 

<Q.2> How to report svc result?
I found many papers just illustrate uncorrected map to show svc results. I wonder if it's better to report corrected results. So,
can I use results of Method II as the small-volume corrected report? And how about the description:

"There was significant activation in this ROI region [MNI coordinates of local maximum: (-2 -28 -8), t = 4.36, p = 0.028]."

I also noticed that some papers did report degrees of freedom in their svc results but others didn't. What does that mean? In my case, at the bottom of the results table, it showed "Degrees of freedom = [1.0, 13.0]. So should I report "[MNI coordinates of local maximum: -2 -28 -8, t(13) = 4.36, p = 0.028]"? But I don't know what does 13 mean.

>>> I personally think reporting the t and p-value is excessive. I would report the corrected p-value. The t-statistic doesn't explicitly translate to the p-value in the case of the corrections.

I would use the sentance:  "We observed significant activation in X (MNI: -2,-28,-8; SVC FWEp=0.028)."

I do appreciate for any help. Thanks.

Mark