No cross-hemisphere tracts in tractify

1

We are working on some diffusion data using tracktify, and building a connectivity matrix using a 200 parcel atlas (schaefer). We’re relatively new to these pipelines.

We are finding that we don’t get very many interhemispheric connections, almost everything is the on the same side.

Here’s what out matrix looks like, its divided right-left as top-bottom half, and sorted by networks.

image
image1920×1920 128 KB

We’ve also been looking at scan-rescan reliability and it appears to be pretty poor, like correlation of 0.4 and below for most edges; here’s a histogram of the correlations or tractify outputs for each connection on scan-rescan (its probably under estimating because its longitudinal).

image
image1920×720 70.4 KB

We have found similar effects in a second data set, i.e. poor scan-rescan and no cross-hemisphere.

Our pipeline is below for reference.

Advice appreciated on
1) why we don’t see cross hemispheric connections
2) if this replicability seems worse than expected
3) if there is maybe soemthing in the pipeline below we could make better in some way?

Pipeline:

Atlas: Schaefer2018_200Parcels_7Networks_order_FSLMNI152_2mm_flirt.nii.gz

Template: MNI152_T1_2mm_brain.nii.gz

## Extracting SSE from b1000s

# Extract b0s and b1000s from dwi

dwiextract --export_grad_fsl dwi_b1000.new_bvecs dwi_b1000.new_bval -fslgrad <bvecs> <bval> -shell 0.0,1000.0 <dwi.nii.gz> dwi_b1000.nii.gz

# Skullstrip the b1000s using bet

bet dwi_b1000.nii.gz dwi_b1000_brain.nii.gz -f 0.50 -m -R

# Get sse from eddy b1000s using dtifit

dtifit -k dwi_b1000.nii.gz -o dtifit_ -m dwi_b1000_brain.nii.gz -r dwi_b1000.new_bvecs -b dwi_b1000.new_bval --save_tensor --sse

## B0s

# Extract the b0s from the dwi

dwiextract -bzero --export_grad_fsl dwi_b0.new_bvecs dwi_b0.new_bval -fslgrad <bvecs> <bval> <dwi.nii.gz> dwi_b0.nii.gz

# Average out the b0s

mrmath -axis 3 dwi_b0.nii.gz mean dwi_b0_avg.nii.gz

# Skullstrip the average b0s using bet

bet dwi_b0_avg.nii.gz dwi_b0_avg_brain.nii.gz -f 0.50 -m -R

## Gray white matter interface

# Generate the five-tissue-type image (5TT) using freesurfer data

5ttgen freesurfer -nocrop <aseg.mgz> 5TT.mif

# Convert the 5TT to gray white matter interface to seed streamlines from

5tt2gmwmi 5TT.mif gmwmi.mif

## Registering atlas to diffusion space

# Register the T1 to diffusion space (t1_flirt)

flirt -in T1_brain.nii.gz -ref dwi_b0_avg_brain.nii.gz -out T1_to_dwi.nii.gz -omat T1_to_dwi.mat -dof 6

# Register the T1 to the template (pre_atlas_flirt)

flirt -in T1_brain.nii.gz -ref <template> -out T1_to_template.nii.gz -omat T1_to_template.mat

# Invert the matrix to get template->T1 matrix

convert_xfm -omat template_to_T1.mat -inverse T1_to_template.mat

# Concatenate the template->T1 and T1->diffusion to get template->diffusion matrix

convert_xfm -omat template_to_dwi.mat -concat T1_to_dwi.mat template_to_T1.mat

# Use the MNI->diffusion matrix to register the atlas to diffusion space (atlas flirt)

flirt -in <atlas.nii.gz> -ref T1_to_dwi.nii.gz -out atlas_to_dwi.nii.gz -omat atlas_to_dwi.mat -applyxfm -init template_to_dwi.mat -interp nearestneighbour

## Getting tracts and filtering

# Generate response function using the multishell 5tt algorithm

dwi2response msmt_5tt -fslgrad <bvecs> <bval> -mask dwi_b0_avg_brain_mask.nii.gz dwi.nii.gz 5TT.mif wm.txt gm.txt csf.txt

# Estimate the fibre orientation distributions (FOD) from the dwi

dwi2fod msmt_csd dwi.nii.gz wm.txt -fslgrad <bvecs> <bval> -mask dwi_b0_avg_brain_mask.nii.gz FOD.mif gm.txt gm.mif csf.txt csf.mif

# Generate tracts from gray white matter interface

tckgen -act 5TT.mif -algorithm iFOD2 -samples 4 -output_seeds out_seeds.nii.gz -seed_gmwmi gmwmi.mif -select 1000000 FOD.mif tracked.tck

# Filter the tracts to match the FOD lobe integrals

tcksift2 tracked.tck FOD.mif prob_weights.txt

## Generating connectivity matrices and outputs

# Generate the connectivity matrix scaled by length

tck2connectome -tck_weights_in prob_weights.txt -keep_unassigned -scale_invnodevol -scale_length -assignment_radial_search 4.000000 -symmetric -zero_diagonal tracked.tck atlas_to_dwi.nii.gz conmat_length_invnodevol.csv

## Other outputs

# Convert the 5TT for output

mrconvert 5TT.mif 5TT.nii.gz

# Convert the gray white matter interface for output

mrconvert gmwmi.mif gmwmi.nii.gz
2 Likes
2

Could you provide screenshots of the outputs of each step visualized using mrview? This will make it a lot easier to pinpoint where things go wrong.

Also, whenever you use a command that uses multiple inputs (e.g. data and mask or data and parcellation), could you visualize them simultaneously (e.g. as main image and transparent overlay) to make sure that they spatially match?

Also, could you visually check whether each registration step was successful (e.g. using overlayed image)?

3

Thanks for the reply, and sorry to delay responding, we needed to find time to generate the images and it was Thanksgiving this weekend in Canada : )

Rather than post a huge number of images on the forim, we made a google doc

https://docs.google.com/document/d/1zNJrpEf_nS4XO4m95AdMDUTAjAv0OEgm2rMs8uepUqY/edit?usp=sharing.

We chose a random subject. There does in this one specific case seem to be a slight BETing issues in the right prefrontal. Otherwise I think it looks OK?

4

Anya able to suggest why we aren’t seeing anything cross hemisphere? We would like to move forward on our analysis

side bad I noticed I called it tracify because that’s what we named our pipeline, its an MRITrixx CSV pipeline seeding from the gray matter interface.

5

Any able to suggest why we aren’t seeing anything cross hemisphere?

Personally I would counter with: On what basis are you expecting a greater propensity of inter-hemispheric connections than what you see? :upside_down_face:

In my own attempt to validate relevant methods I found this chapter claiming based on post mortem dissection that around 2% of all WM fibres traverse the corpus callosum; and with post-SIFT connectomes this is approximately what I observed in my empirical data. I suspect many pipelines over-reconstruct these trajectories.

Nevertheless, if your number is even lower, at least you know where to look: you can check for adequate volumetric coverage of the corpus callosum and whether the streamlines density is unusually small there.