Hi there,

how "far" are you from reaching statistical significance? are your results corrected for multiple comparisons? if so, you can try running randomise with the --uncorrp flag and check your (uncorrected) p-values again

which pipeline did you follow to process your images? have you eyeballed your FA images for a visual quality-check?

also, have you double-checked your design matrix and contrast?

equally, with sample sizes like yours (N=15) you may benefit from smoothing your variance as suggested in the FSLwiki (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/UserGuide)

on another note, it is also possible that the effect of training may relate to changes in tract density, or even some underlying fibre re-wiring. Both situations can reflect a decrease in FA wrongly suggesting changes in WM microstructure. This is, diffusion could be less constrained in long or light-packed tracts or territories with many crossing fibres.

maybe some of the following papers would help in interpreting your results!

Article Jones DK, Knosche TR, Turner R. White matter integrity, fibe...

Article Multimodal Voxel-Based Meta-Analysis of White Matter Abnorma...

Article Zatorre RJ, Fields RD, Johansen-Berg H. Plasticity in gray a...

Hello Jonatan,

Thanks for your replay

I followed the standard pipeline from this website: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/TBSS/UserGuide

what do you mean by a visual quality-check? I checked them by my eyes before I start the randomize

have you double-checked your design matrix and contrast?

what I did is that I split my all_FA_skeletonised which has 30 images ( 15 before and 15 after) into individual skeletons

then, used fslmaths to subtract images from each other and I got only 15 skeletons and I merge them into one 4D skeleton and run one sample T test the reason of doing this because I have one group not two!

Finally, I used this commond :

randomise -i all_FA_skeletonised -o OneSampT -1 -v 5 -m bin_mean_FA_skeleton_mask.nii.gz - -T2

I hope you can help me more

Thanks again

Manal

hi Manal,

I just re-read what you wrote and had to edit my message because you've already tried some of the things that I've suggested.

You could try with a single-group paired difference approach (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Single-Group_Paired_Difference_.28Paired_T-Test.29).

On a different note, it could worth the effort to test for other DTI metrics (MD, RD, AD... check the tbss_non_FA script in FSL). It is unlikely, but maybe the FA increase is subtle but changes in other scalars not (i.e., decrease in RD indirectly suggesting an increase in myelination).

Let me know if this have worked for you!

J

Hello J,

I started looked to MD data using reg -n this time but it seems it will take more time then I thought would be !

is there any way to make it work faster ?

the reson why I did this because reg -T with MD images is so bad every time I checked all MD files found more then two were failed

do you have any suggestion for processing MD L2 RD images ?

Thanks

Manal

Hey Manal,

you don't have to register again all your images for complementary DTI metrics. You only need to fit those scalars to the already normalized average FA skeleton with the tbss_non_FA script.

Briefly, copy all your subject's MD images to a new folder named MD in the tbss directory.

NOTE: Just make sure to name these MD files exactly as the images in the origdata folder where your original FA maps are, even if they're not FA volumes (otherwise the script will crash).

now from the tbss directory type in a terminal tbss_non_FA MD

this will take 3 to 5 minutes to process. In the stats folder you'll find the all_MD_skeletonised.nii.gz to run your statistics (use the mean_FA_skeleton_mask.nii.gz regardless of the type of image, all your supplementary metrics are now projected to this skeleton).

Concerning the other metrics. AD is the same as the L1 images. RD is the average of L2 and L3 volumes (check this to aim interpretation: http://www.diffusion-imaging.com/2013/01/relation-between-neural-microstructure.html)

repeat the abovementioned step for the other metrics (tbss_non_FA AD, tbss_non_FA RD).

Best,

J

Hello J ,

I did this and the registration did not work as you can see in the attachment screen shot that why this time I decided to work with MD files rather then FA !

Thanks

yes, that screenshot doesn't look very good...

however, you should try the tbss_non_FA script as I suggested, let me know if that worked fine for you!

J

Hello J ,

Do you any idea why my FA maps look like this (upside down) !

Thanks

Manal

I have no idea why your images are upside-down, but it's something I've seen happening latelty to other people!

which software have you used to convert your images? are your images already flipped in the raw DWI volumes, or they turn upside-down once you fed them to FSL?

if you're using dcm2nii, you could try with MRIconvert (https://lcni.uoregon.edu/downloads/mriconvert).

The same thing happened to me in a lesser extent (only a few subjects), and this software did the trick for me.

Best,

J

Hello J,

I used this command :

unpacksdcmdir -src -targ -fsfast -seqcfg

Thanks

Manal

Hello J,

Cloud you please have a look at this :

I run tbss_2_reg -n

then, tbss_3_postreg -S

and this is what I got!!

Thanks

Manal

Manal,

perhaps you haven't readed my last reply,

this "upside-down" thing is already happening to you with the raw volumes? or it is occuring once transformed the raw dicoms to NIFTI with fsl?

please try to reconstruct a single subject either with the Chris Roden's tool dcm2nii or the MRIconvert software.

Hello J,

This happened with only one data set Not with other of my data , even though they all looked upside-down in raw !

that is why I did not try a new method for reconstruct because it works with some data but Not all unfortunate :(

what do you think the reason for this?

Thanks

Manal