Ques about Automatic Segmentation and Incomplete Whole Brain Tractography

[Luie Wang]

Dear AFQ community,

I'm new to diffusion imaging and wish to use AFQ in conjunction with MRtrix3 to analyze cingulum cingulate and cingulum hippocampus in patients vs. control. I apologize in advance if my questions are very basic. 

I'm confused about the automatic segmentation of fiber groups in AFQ. From AFQ wiki page, it seems like the dMRI images (after aligned to ac-pc T1 image) would still be in their native/individual subject space. For AFQ_SegmentFiberGroups to work, do we have to create ROIs for each participant specifically? Or does AFQ saves us from the manual ROI selection step and automatically extracts the white matter tracts from ROIs saved in AFQ itself? Are registration (aligning the individual dMRI images) required for this to work? 

My dataset is already preprocessed via FSL. I created dt6.mat from dtiInit and tried to generate whole brain tractography in test mode for 1 participant. However, the resulting tractography looks incomplete. As shown in the attached diagram, the tracts are missing in the anterior part of the brain. I don't know what's the cause of this weird result. Could it be some problem with the ac-pc alignment (since I did it manually via vistasoft)? Also, during the process of generating dt6.mat, it shows the phase and frequency dimensions of my image is not set correctly in the Nifti header, is this something I can ignore or could it be a reason leading to my weird outcome? 

Any advice/suggestion is greatly appreciated! Thank you so much in advance for all your helps!

Best,

Luie

[Andrei Vakhtin]

Greetings AFQ Experts,

I am curious if any ideas or solutions regarding this came forward? We are having a very similar incomplete whole-brain tractography result on GE Discovery data (ADNI2). Interestingly, the Siemens runs fine with the dwParams.rotateBvecsWithCanXform = 1. With this parameter set back to 0 for GE data, the image from Luie's post is reflective of what we get. 

Thanks for your time!

Andrei

[Luie Wang]

Hi, 

I eventually solved the issue by flipping the bvecs over the x-axis. My data were required from Siemens, and the fiber looks incomplete even with dwParams. rotateBvecsWithCanXform =1. So I tried flipping the bvecs over x-axis and it works out for me. Hope this may solve your problem.

Best,

Luie

[Maria Economou]

Hi Luie,

May I ask exactly how you managed to flip the bvecs over the x-plane? I am struggling to find the solution for this somewhere!

Many thanks,

Maria

[Jason Yeatman]

In many cases you can just multiply the X gradient direction by negative 1. There are also some different options in dtiInitiParams for how to deal with the bvecs across different scanners

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/automated-fiber-quantification-afq/281b711f-9b84-45fb-900e-64cafd246fc7n%40googlegroups.com.

[luie...@gmail.com]

Hi Maria, 

I multiplied the x-gradient in bvecs file by -1 and did this using matlab. If you open the bvecs file in matlab, you can see it's separated into three rows, each representing one gradient direction. 

Hope this helps. 

Luie

[Maria Economou]

Thanks both for the tips!