FWE Corrected Results vs Uncorrected Results Display Discrepancy

[Chloe Cheung]

Dear SwE Experts,

I am currently using SwE to look at structural T1 scans from a longitudinal study. However, I have noticed clusters when displaying results with FWE correction = 0.05 but that do not appear when displaying unthresholded results. Are there any steps I can take to resolve this issue? Please let me know if any additional information is required.

Thanks in advance for your help,

Chloe

[Thomas Nichols]

Dear Chloe,

That is a bit strange, but not inconsistent.  Your "Corrected" figure is based on cluster forming threshold of 0.01 ("Z>3.09") in the footer, while your uncorrected figure is based on p<0.001 and you've only used 100 bootstrap samples and thus the smallest possible p-value is 1/100=0.01.  Try *not* using the Non-Parametric Wild Bootstrap to get thresholding that is based on parametric p-values that can of course be smaller than 0.01 depending on the data.

-Tom

--
You received this message because you are subscribed to the Google Groups "SwE-Toolbox Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to swe-toolbox...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/swe-toolbox/6387dd1f-bdd3-4ea4-91c6-87468ffcf749n%40googlegroups.com.

[Chloe Cheung]

Dear Tom,

Thank you so much for your time and help! Your suggestion worked and and the unthresholded clusters are now visible.

I have a few follow up questions:

1. To confirm, if I increase the number of bootstraps past a certain point (EG: 1000 bootstraps), would I then see clusters in the unthresholded results? If so, what is the minimum number of bootstraps you would recommend?
2. The model is currently using explicit masking with ICV.nii, but some clusters are present outside the brain even if a different mask such as EPI.nii is used. Are there any changes I can make to get rid of them?

Thanks again for your time,

Chloe

[Thomas Nichols]

Dear Chloe,

I've never heard of an ICV.nii... ICV is intracranial volume and is a single number, not an image.  What is this image?

Anyway, can I refer you to Ridgway et al (2009) which stresses the need for an explicit mask.  That you're using now is great, but clearly it is not stringent enough. While that paper proposes some fancy methods, you can use something really simple: Take the mean of your VBM inputs and threshold them at, say 0.1... in any event, you can look at the mask that gets created by mean>0.1 and see if it is sensible.. if it includes too many non-brain voxels, increase the threshold.

Let me know if this makes sense.

-Tom 

Ridgway, G. R., Omar, R., Ourselin, S., Hill, D. L. G., Warren, J. D., & Fox, N. C. (2009). Issues with threshold masking in voxel-based morphometry of atrophied brains. NeuroImage, 44(1), 99–111. https://doi.org/10.1016/j.neuroimage.2008.08.045

To view this discussion on the web visit https://groups.google.com/d/msgid/swe-toolbox/65e7ad6b-5e96-479b-b6be-67d0fb014be7n%40googlegroups.com.

[Chloe Cheung]

Dear Tom,

Thanks for getting back to me so quickly! I will take a look at the Ridgway 2009 paper.

Sorry about the confusion, the ICV.nii file is referring to the spm12/tpm/mask_ICV.nii file found in the spm12/tpm/ folder.

To confirm, I would 1. create a mean or average image from all my VBM images and then 2. apply an intensity threshold of 0.1 to the mean image and 3. create a new explicit mask from the voxels above that 0.1 threshold?

Thanks again,

Chloe

[Thomas Nichols]

Dear Chloe,

Sorry about the confusion, the ICV.nii file is referring to the spm12/tpm/mask_ICV.nii file found in the spm12/tpm/ folder.

Ah... but I can see that is a rather generous mask... look at the great amount of in-mask voxels that are far from brain tissue on the midsagittal plane...

... not to mention lots of white matter on the interior of the brain.

 

To confirm, I would 1. create a mean or average image from all my VBM images and then 2. apply an intensity threshold of 0.1 to the mean image and 3. create a new explicit mask from the voxels above that 0.1 threshold?

Exactly!

-Tom

 

To view this discussion on the web visit https://groups.google.com/d/msgid/swe-toolbox/dcdfaa8a-3065-4c23-a49c-b36fc34c9eb3n%40googlegroups.com.

[Chloe Cheung]

Dear Tom,

I was able to create 0.1 intensity and 0.2 intensity thresholded explicit masks derived from the mean of the VBM smwp1r* files, and it looks like they're working!

Seems like the clusters that were appearing outside the brain were actually within the brain all along.

Thank you so much for all your help and support!

Chloe

[Thomas Nichols]

Super!

To view this discussion on the web visit https://groups.google.com/d/msgid/swe-toolbox/f13d2a84-f86a-40da-b2cc-179f0a2369f9n%40googlegroups.com.