Dear Evan,
What you need to do with fslmeants is create a mask that only contains the voxels that you definitely want included. There is no option for excluding zero voxels with fslmeants.
I believe that the best option is to only look at the voxels on the skeleton, and if you want this broken down into regions (as defined by the JHU atlas) then you need to make masks that are the intersection of the skeleton and JHU regions. So take the JHU masks, and multiply them by a skeleton mask (if you don't have one then make one by taking the a skeleton image and binarising it with fslmaths). Once you've multiplied the skeleton mask with a JHU-region-mask then use the result as the mask in the -m call to fslmeants.
This should then produce results that are equivalent to the second fslstats call that you made - using the skeletonised input, the Peduncle mask and the -M option (since that only looks at non-zero voxels within the Peduncle mask). The other calls you made to fslstats either include zero voxels within the Peduncle mask (option 1), or included lots of non-skeleton voxels (options 3 and 4).
I hope this helps.
All the best,
Mark
On 7 Feb 2014, at 14:39, Evan Stone <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi there,
I used fslmaths to create masks from the JHU White Matter Tractography atlas using -thr, renamed each of the 48 ROIs to the appropriate title, and used it with fslmeants after the -m command as explicit masks for the all_FA file. It doesn't rub me the right way that the data i get from this is equivalent to as if i told fslstats to include nonzero voxels by using -m; what zero-value voxels would be in this? Also, using the skeletonised tract with this mask derived from a normal brain consistently produces values waaaay below expected...
On Feb 7, 2014 11:45 AM, "Mark Jenkinson" <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi,
I completely agree with this and just want to add that with fslmeants you need to use the -m option to specify an explicit mask, rather than have it automatically select non-zero voxels. Creating such a mask, for the non-zero voxels, is very easy with fslmaths.
All the best,
Mark
On 7 Feb 2014, at 07:18, "Watson, Christopher" <[log in to unmask]<mailto:[log in to unmask]>> wrote:
> "all_FA_skeletonised" is going to only contain the voxels that are part of the FA skeleton (see Steve Smith's 2006 NeuroImage paper, and the wiki), which are the data that would be included in a TBSS analysis. I won't say if one way or the other is "better" (because I don't know the answer), but I would choose the skeletonised data for my own study. For an ROI such as the cerebellar peduncle, it might turn out that there are too few skeletonised voxels to give you reliable data (depends on your particular group). In that case, if the raw data are good enough (e.g. not too many partial volume effects), you may wish to use "all_FA". But inspecting the raw data *and* the processed data is important.
>
> Regarding "fslmeants" vs. "fslstats", and whether or not to include nonzero voxels: I would say you should always exclude zero voxels, because they will just increase the N, if you're taking the mean FA of a given region. But those voxels won't contribute, and should be excluded. I don't know if the commands do something different in this case; I think they should give the same result.
>
> ________________________________________
> From: FSL - FMRIB's Software Library [[log in to unmask]<mailto:[log in to unmask]>] on behalf of Evan Stone [[log in to unmask]<mailto:[log in to unmask]>]
> Sent: Friday, February 07, 2014 1:38 AM
> To: [log in to unmask]<mailto:[log in to unmask]>
> Subject: Re: [FSL] Data Extraction
>
> Yes thank you, but why would this be different from using allFA? Is this the right way to gather the data, or is there a better way? And how would i use fslmeants for nonzero voxels?
