 |
 |
Hi Gwenaelle, Thank you very much for the helpful information in your last reply. I?ve implemented your suggestions and have gotten some nice results. I am, however, running into some difficulties with the mean FA for each subject in the significantly different regions because the odd distribution of these regions. In my results, there are only two clusters, one with over 12,000 voxels and with other with just 100 voxels. I thought a good approach might be the following: 1) Create the text file listing all significant voxels with corresponding FA (i.e. all_voxels_significant.txt) separately for all patients. 2) For each subject individually, group the all significant voxels according to one of the 20 white matter structures listed in the JHU White-Matter Tractography Atlas. The label assigned to the voxel would dependent on which has the highest probability. 3) Compute mean FA individually for each subject according to white matter structures. Is there an automated or other efficient way to do this such as feeding in the coordinates of all_voxels_significant.txt into the JHU White-Matter Tractography Atlas and getting labels for these voxels? I could then bin them manually according to white matter structure and compute mean FA values. Better yet, does a program within FSL exist to perform all of these steps with minimal manual intervention? As always, your input is greatly appreciated! Regards, Roger Quoting Gwenaëlle DOUAUD <[log in to unmask]>: > Hi Roger, > >> 1. When I apply the binary mask (thresholded at 0.95) to >> mean_FA to obtain the >> text file all_voxels_significant.txt: > > Each column represents a voxel: > MNI_x1 MNI_x2 > MNI_y1 MNI_y2 > MNI_z1 MNI_z2 > meanFA1 meanFA2 etc. > >> b. Since the input file in >> fslmeants is mean_FA, do the FA values in the >> lower half of the text file represent mean values for the >> entire sample >> (controls and patients mixed together)? > > Yes, as you asked in your email for the mean FA without specifying > that you wanted it split by groups. > > If so, how >> can I do this so I get >> means of each group separately? > > Yes, just do an fslmaths -Tmean on your 4D only containing the > controls (you can just use fslroi on the all_FA_skeletonised to do > so), then only the patients and repeat the fslmeants --showall for > each of the mean_FA created. > Having said that, I have the feeling that what you might want is not > *all* significant voxels, but the mean FA for each subject in the > significant regions. If that so, then simply do: > > fslmeants -i all_FA_skeletonised -m mymask -o my_significant_FA_results.txt > >> 2. Creating ROI's from the structures in the JHU >> White-Matter Tractography >> Atlas, a probabilistic atlas: >> >> a. What is the acceptable >> threshold range when creating the mask using >> fslmaths? > > Hmmm, as you found out in the archives (thanks for pointing at it for > Markus btw), there is no good answer as this is always arbitrary. Now > again, 5 to 20% seems reasonable. > >> b. I understand this setting is >> arbitrary. Thus, is there any >> disadvantage >> or contraindication to use an even lower threshold (say >> 1%)? > > Not really, but I am not sure why you would want to go that low (the > more conservative you are, the surer you are to be in that specific > tract)... Anyway, I don't think choosing 1% or 5% would make any > difference after intersecting with the white matter skeleton. > >> 3. When intersecting each ROI created from the >> structures in the JHU >> White-Matter Tractography Atlas with my mask of significant >> results: >> >> a. What is the chance that >> identified voxel(s) present in one mask, be >> present in many others? > > By setting up the threshold higher (see above), you reduce this > possibility. Now if your results are right in the middle of some > crossing fibres, this is then meaningful to get the results in two > different masks. > >> b. What is the best way to use the >> mask resulting from the intersection >> (i.e. mymask_unc in your example below), to obtain mean ROI >> FA on each of the >> subject individually? > > Aha, answered this one already above. > > Cheers, > Gwenaelle > > > >