DTI-TK with TBSS - non-FA metrics

[Richard Parker]

Hi Gary, 

This morning I tried out your pipeline for integrating DTI-TK's tensor registration with TBSS, and the results look great. I now have a skeleton containing all of my subject's FA data. I was wondering if you could clarify how I can repeat the process, but using non-FA data?

At a guess, I think that I would keep the files 'mean_FA' and 'mean_FA_skeleton' that were created when I ran TBSS on my FA maps. I would then generate maps of my metric of interest (e.g, RD) and merge them together, naming the file as 'all_FA' to keep in line FSL's naming convention. I would then place all these files a new directory, and run 'tbss_4_prestats' as usual, resulting in a skeletonised map of radial diffusivity for my sample (although the file would be called 'all_FA_skeletonised'). 

Does this sound about right to you? What I am unsure of is whether to keep the versions of 'mean_FA' and 'mean_FA_skeleton' that were created when I ran through the pipeline with the FA data, or whether to generate new versions, based upon my new metric of interest. 

Best wishes,

Richard Parker

King's College London

[Hui Zhang]

Hi Richard,

Thanks for trying DTI-TK.  I'm copying Inge who has managed to use non-FA data for TBSS.

Inge, can you help Richard out with this?  I am sure it will be useful for a lot of other DTI-TK / TBSS users.

Gary

--
 
---
You received this message because you are subscribed to the Google Groups "DTI-TK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dtitk+un...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Timmers Inge (PSYCHOLOGY)]

Dear Richard,

 

Your guess is quite correct. In our analyses we also used the original mean_FA and mean_FA_skeleton for the final part of TBSS on non-FA data. Further, we kept the original distance maps created during the FA pipeline (mean_FA_skeleton_mask_dst), because the creation of the distance maps failed using other non-FA data (you can read more about the details of the distance maps in the paper).

 

The key then is to use the TBSS_non_FA function, but then without the registration procedure. Digging into the code will teach you that they use the following line for applying the distance maps to project to non-FA data onto the skeleton:

 

“fsl5.0-tbss_skeleton -i <input image> -p <skeleton_threshold> <distance map> <search_rule_mask> <original 4D_data> <porjected_4D data name> -a <alternative_4D data> [-s <alternative_skeleton>]”

 

in which you can keep mean_FA as input, original distance maps, the standard search_rule_mask “/usr/share/fsl/5.0/data/standard/LowerCingulum_1mm”, original all_FA data, original mean_FA map, and then add your new non_FA data (the all_RD e.g.).

 

This should do the trick.

 

Good luck,

 

Best,

Inge

[Manuel Blesa Cábez]

Dear Richard,

Sorry for refresh this topic, but I'm stuck doing the same thing (creating my nonFA_skeletonised file)

I did this:

“fsl5.0-tbss_skeleton -i <input image> -p <skeleton_threshold> <distance map> <search_rule_mask> <original 4D_data> <porjected_4D data name> -a <alternative_4D data> [-s <alternative_skeleton>]”

fsl5.0-tbss_skeleton -i mean_FA -p $thresh mean_FA_skeleton_mask_dst ${FSLDIR}/data/standard/LowerCingulum_1mm all_FA mean_FA all_MD_skeletonised -a all_MD

Adding all_MD_skeletonised to change the name to the new skeletonised file, it's the same?

And what's the role of the threshold? because I change the value and the output file is the same.

Thanks in advance

Regards,

Manuel Blesa

[Hui Zhang]

Dear Manuel,

Have you seen Inge's response to Richard in this thread?

Gary

--

---
You received this message because you are subscribed to the Google Groups "DTI-TK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dtitk+un...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Manuel Blesa Cábez]

Hi Gary,

Thanks for the answer, I run the command that Inge says, and it creates a file called -a, I modified a little bit the command to obtain the skeletonised file,and I would like to know if my changes are correct (all_MD_skeletonised before -a).

And I don't know what's the role of the threshold, because I change this parameter and the output is the same.

Thanks in advance

Regards,

Manuel Blesa

[Hui Zhang]

Hmmm.  There could be a typo or something.  I am copying Inge.  She may be able to comment.

Gary

--

---
You received this message because you are subscribed to the Google Groups "DTI-TK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dtitk+un...@googlegroups.com.

[Inge Timmers]

Dear Manuel,

First, the threshold is the same threshold as FSL TBSS uses in the final step 4, so to threshold the FA skeleton image. 0.2 is given as an example, but you should find out what works well for you data. Indeed, changing the parameter might not cause massive changes.

As far as I can judge from the code you described above, you have one entry too many ("mean_FA"). The code is also not supposed to create a file called "-a", but -a should indicate the entry of the alternative dataset ("all_MD" in your case). This probably occurred because of the extra, unexpected, entry. 

Also, if you want to give an alternative skeleton (i.e., the FA skeleton you created, and I suggest you do as you are also using the original distance maps), you should add ("-s mean_FA_skeleton" or whatever name the FA skeleton has).

Could you check again, using the adapted code below? Now it should work properly.

"fsl5.0-tbss_skeleton -i mean_FA -p $thresh mean_FA_skeleton_mask_dst ${FSLDIR}/data/standard/LowerCingulum_1mm all_FA all_MD_skeletonised -a all_MD -s mean_FA_skeleton"

Good luck!

Inge

[Sila Genc]

Hi all,

I had the same problem, and I wonder if it is even worth doing the skeleton projection step seeing as though some papers quote that it is redundant, and could cause overfitting of the white matter skeleton, particularly since DTI-TK is supposed to improve registrations. I have used the mean_FA_skeleton_mask as a mask for my FA, MD, RD, and AxD maps - skipping the skeleton projection - in order to overcome issues with TBSS integration.

Do you, Gary and Inge, think this is appropriate?

Warm regards,

Sila

[Hui Zhang]

Hi Sila,

It is an unfortunate misconception that the skeleton projection step is just for mitigating the effect of poor registration.  The step actually is important for more than that reason alone.  In particular, it provides a means to reduce the dimensionality of the problem, namely from 3D volume to 2D skeletons.  This will reduce the number of multiple comparisons and boost the sensitivity.

Can you look at Inge's responses and follow her suggestions?  If you still have issues, please document precisely the commands you used in your follow-up message.  That is the only way that others can help you with this.

Gary

[Hui Zhang]

Hi all,

Let me try to clarify how to use tbss_skeleton for non_FA metrics.

Using MD as an example, the command should read like this:

tbss_skeleton   -i mean_FA    -p $thresh    mean_FA_skeleton_mask_dst    ${FSLDIR}/data/standard/LowerCingulum_1mm all_FA  all_MD_skeletonised -a all_MD 

I have highlighted each of the seven arguments by alternating the font styles.

Hope this helps.

Gary

[Sila Genc]

Thanks Gary that is helpful - all working now.

Kind regards,

Sila