How to get 3D tensor model with a gradient table different for each slice?

[Pedro-Filipe]

Hello,

WARNING Noob question alert.

I have some multislice diffusion weighted data. I am trying to calculate a tensor model with tenmodel = dti.TensorModel(gtab) to ultimately perform some tractography.

But I don't have an identical list of bvals and bvecs for each slice, so my gtab will be different for each slice. How can I get my 3D tenmodel?

Many thanks for any help.

[pjsjongsung]

Hello, could you explain your situation in more detail? What are the shapes of your volume, bvals and bvecs?
From what I hear, you could calculate the tensor per slice by initializing a dti.TensorModel every slice and running fit on them.

It would be a little bit inconvenient, since every time you want, for example, a FA image, you would have to run a loop through the slices. However, that is the easiest solution I can think about at the moment.

[Pedro-Filipe]

Hi @pjsjongsung,

Thanks for your reply.

I have N slices. For each slice a series of 2D diffusion weighted images. But the number of images, values of bvals and bvecs varies from slice to slice.

I was hoping to do the following steps (simplified code):

tenmodel = dti.TensorModel(gtab)
tensor_peaks = peaks_from_model(model=tenmodel,...)
streamlines_generator = LocalTracking(tensor_peaks, ...)

No problem at all in calculating the tensor model per slice. But at some point I need to stack my data in 3D. I am not sure what is the best way to do that. Do I need to manually stack all the numpy arrays inside the variable tensor_peaks?

[pjsjongsung]

Yes that sounds about right. You can try concatenating the slice dimension most of the variables (I think peak_indices, peak_values, peak_dirs, gfa, qa and shm_coeff) from tensor_peaks, and then pass it to local tracking. Let us know if it works.

[Pedro-Filipe]

@pjsjongsung It worked!
Manually concatenated the arrays and then passed it on to LocalTracking without a problem.

Many thanks for the help.