Thanks!

____________________
Mark Wagshul, PhD
Associate Professor
Gruss Magnetic Resonance Research Center
Albert Einstein College of Medicine
Bronx, NY 10461

Ph: 718-430-4011
FAX: 718-430-3399
Email: [log in to unmask]
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Anderson M. Winkler [[log in to unmask]]
Sent: Sunday, July 30, 2017 4:09 PM
To: [log in to unmask]
Subject: Re: [FSL] Reporting results from "cluster"

Hi Mark,

Please see below:


On 27 July 2017 at 18:39, Mark Wagshul <[log in to unmask]> wrote:
Dear FSL experts,

I am using randomise to extract areas of FA differences between two groups (~ 60 subjects per group), controlling for the usual covariates (age, gender, education, handedness).  Using TFCE, I find a small cluster of significance at p < 0.05 (139 voxels).  A few questions:

1) How do I get the minimum FWE-corrected cluster size, or is this not a relevant number for TFCE?

It isn't relevant for TFCE.
 

2) I would also like to report clusters seen at a lower threshold.  The (yes, clearly subjective) reason for doing this is a clear bilateral pattern in the data which I can see when looking at the unthresheld data, e.g. if I look at p < 0.3, there are clearly two areas of significance, one centered around the original (p < 0.05) cluster and the other one on the contralateral side.  The anatomical symmetry of the signal gives me some confidence that these are real, but I don't have enough statistical power to detect the contralateral signal at p < 0.05.  So, do you have a recommendation as to how to report these results?  I could simply present the unthresheld maps which show this symmetry, but I'm wondering if there is an objective way to quantify the bilateral nature of the results.  Can I report clusters at a lower statistical threshold, and if so, what would be a reasonable threshold to use?

Presenting the unthresholded map is probably the more appropriate. Any other number tweaked to resemble a significant p-value will be misleading. The results aren't significant. That's it. It's find to present the unthresholded map.

 

3) If I can do what I'm proposing above, namely report clusters at some lower threshold, should all clusters be reported?  When I do this, e.g. for p < 0.3, I get some very small clusters (< 10 voxels).  Should I be reporting even these tiny clusters, or is there a way to objectively set a cluster size threshold and not report those below this threshold?

It doesn't seem a good idea to me.
 

4) A related question - I would like to look at the significance (or non-significance) of differences in other DTI parameters within these clusters (e.g. RD, as well as NODDI-based parameters).  So, do you think it is valid to choose a lower statistical threshold where I can detect bilateral clusters from which to extract these other parameter values across my subject groups?

Not valid because these other measures aren't independent from those used to define the clusters.

Hope this helps.

All the best,

Anderson

 

Thanks for any advice you can provide,

Mark Wagshul