Cluster thresholding after 3dLME

AFNI Message Board
1

Hi there,

I know this topic has been covered before but I’m hoping for some help to make sure I understand all options for cluster thresholding after running 3dLME and/or 3dMVM.

So far, I have tried outputting the residuals then running:
3dFWHMx -ACF NULL -mask mask.nii -input residuals.nii -unif
3dClustSim -acf 0.34947 4.38785 11.4259 -mask mask.nii

(The mask is just a binary whole brain mask).

However, the resulting tables suggest that the clusters will need to be quite large to meet a corrected .05 threshold (k > ~90 voxels with uncorrected pthr=0.001). Two questions:

  1. Outside of choosing an arbitrary cluster size threshold, is this the only/(correct?) way to threshold results by cluster extent from 3dLME/3dMVM? I looked at the etac slides but wasn’t sure if there is a way to try this approach with results from 3dLME.

  2. Do clusters this large seem reasonable or is there potentially an error in my approach? We lose a fair amount of activity in biologically interesting areas this way compared to a simple pthr=0.001, cluster k = 10 or 15 voxels.

Thank you for your help!
Becky

2

Becky,

Outside of choosing an arbitrary cluster size threshold, is this the only/(correct?) way to threshold results by cluster extent from 3dLME/3dMVM?

Determining a minimum cluster size is pretty much like defining where the sea level is (locations of moon and sun, global warming, etc.). So, correctness is likely in the eye of the beholder.

Do clusters this large seem reasonable or is there potentially an error in my approach? We lose a fair amount of activity in biologically interesting
areas this way compared to a simple pthr=0.001, cluster k = 10 or 15 voxels.

The sacrifice of small regions is a heavy price that the common practice of multiple testing correction in the field has to pay. In addition, artificially dichotomizing the statistical evidence is another endemic problem. See more discussions here:

https://www.sciencedirect.com/science/article/pii/S1053811919309115
https://afni.nimh.nih.gov/afni/community/board/read.php?1,160410,160414#msg-160414

3

Hi Gang,

Thanks for the response. I’ve been reading your comments on the other thread about issues with treating the corrected p < 0.05 as a binary pass/fail. Your idea of “highlighting, not hiding”, is especially helpful in framing how to consider and report these results (ie. looking at the broader range of thresholds).

Becky