How to generate white matter tractography?

[sunitaiitb]

Dear DIPY contributors,

We are trying to generate tractography map of brain. DIPY is the library used for this. The instruction we are getting is very confusing. Sometimes DTI is used and sometimes DWI is used. We could understand that .bvac and .bval files generated from DWI image is used for the generation of fiber tracts.

From the web, we find that DTI should have four dimensions however, the DTI file we are having has only three dimensions. How to add the fourth dimension in DTI FA image which provide information about the magnitude and direction of the diffusion?

Which method is correct for the generation of white matter mask "median_otsu" or "tissue_classification"?

Could you please guide us on how to generate the tractography image.

We will highly appreciate your help.

Thank you so much.

Best regards,
Sunita

[pjsjongsung]

It seems you are looking at the FA image, which is a processed image from the DTI, thus being a 3 dimensional image. There must be some volume you feed to your function creating a FA image.

For generating white matter mask, if it is not a skull stripped image, you want to do median_otsu first and then tissue_classification. If the brain is already extracted, you can just to tissue classification.

We have steps generating tracts in our tutorials. Hopefully they would be helpful, if not let us know.

[sunitaiitb]

Dear Jong,

The dataset is from TCIA site which provides incomplete information. For a beginner it is really confusing.

Finally, we got a dataset with DWI in 4D and T1W in 3D. The dimension of DWI is 128, 128, 88, 67 and T1W is 176, 240, 256. The WM mask should match with DWI dimension. For that do we have to coregister and resample T1W to match 128, 128, 88 dimension before extracting WM mask?

or

Is it possible to generate WM mask from DWI image?
We used median_otsu to get the skull stripped brain mask then used DWI image and mask in dti.tensormodel to get FA image.
Now can we get the WM mask from FA image?

Thank you so much.

Best regards,
Sunita

[pjsjongsung]

While it is best if you have a calculate WM mask, it is also possible to get a crude mask by thresholding the FA image. This tutorial could provide some information on stopping criterions.

[sunitaiitb]

Dear Jong,

Thanks for your guidance.
At least now we could generate the tracts. But to generate meaningful tracts we need understand the functions. The tractograph looks weird may be because of wrong selection of seed.

We loaded the DWI data with bvec and bval values as follows.

data, affine, img = load_nifti(dwi_, return_img=True)
bvals, bvecs = read_bvals_bvecs(val_, vec_)
gtab = gradient_table(bvals, bvecs) 

Then we ran median_otsu for brain extraction of DWI.
The shape of DWI is 128,128,88,67. What does vol_idx[3] mean here?

dwi_data, dwi_mask = median_otsu(data, vol_idx=[3])

We used TensorModel() to generate the FA. The mask we got from median_ostu() is in boolean form.

dti_model = TensorModel(gtab)
dti_fit = dti_model.fit(dwi_data, mask=dwi_mask)
FA = fractional_anisotropy(dti_fit.evals)

When we visualize with 0.2 threshold we got following image. Do you think this image is acceptable for serving as white matter (WM) mask?

We are taking FA as white matter. Now how to set the seed mask? We tried setting greater than 0.6 threshold as the seed mask. With this threshold the image appeared as follows.

while_matter=FA.copy()
seed_mask = (labels == 2)
seed_mask[white_matter < 0.5] = 0
seeds = utils.seeds_from_mask(seed_mask, affine, density=2)

After setting the seed we ran the following code. But it ran for 50% and stopped. Is there anything to change here?

response, ratio = auto_response_ssst(gtab, data, roi_radii=10, fa_thr=0.7)
csd_model = ConstrainedSphericalDeconvModel(gtab, response)
csd_fit = csd_model.fit(data, mask=white_matter)

dg = DeterministicMaximumDirectionGetter.from_shcoeff(csd_fit.shm_coeff,
                                                      max_angle=30.,
                                                      sphere=default_sphere) 

After that we ran the local tracking and visualize the tractograph.

interactive=True
streamline_generator = LocalTracking(dg,
                                     stopping_criterion,
                                     seeds,
                                     affine,
                                     step_size=.5,
                                     return_all=True)
streamlines = Streamlines(streamline_generator)
sft = StatefulTractogram(streamlines, img, Space.RASMM)
if has_fury:
    scene = window.Scene()
    scene.add(actor.line(streamlines, colormap.line_colors(streamlines)))
    window.record(scene, out_path='tractogram_deterministic_thresh_all.png',
                  size=(800, 800))
    if interactive:
        window.show(scene)

After this we are getting this image.

In this image the threshold values greater than 0.6 is used as seed for generation of fiber tracts. How to decide the seed? How to get the correct tractograph? How to generate corticospinal tract, corona radiata, corpus collasum etc?

Thanks a lot.
Sunita

[Sunita2023]

Dear Jong,

We have understood on how to generate white matter tracts. Thanks DIPY team for the wonderful tutorials. We generated the full brain streamlines and extracted the bundle of interest by comparing the recobundles.

Thank you all DIPY developers.

A little more guidance from DIPY team would have been a great help.

Kindly please provide some guidance when we asked question.

Thank you so much.
Sunita

[pjsjongsung]

Glad it worked! :) I'm sorry I missed the previous question. Please feel free to add a comment or ping me as a reminder if there is no response in the future.