Add polynomial function to dummy_segmentation

[valosekj]

This PR aims to address #2678 (creation of synthetic interleaved DWI data).
So far, I added polynomial function to dummy_segmentation to simulate slicewise motion:

TODO:

• allow creation of 4D dataset - possible solution: create function which will call dummy_segmentation n-times with various factor of polynomial function and then merge individual 3D data into 4D dataset

[valosekj]

dummy_4d function allows to create 4D data with different slicewise shift across individual volumes and can be tested for example by following code:

from spinalcordtoolbox.testing.create_test_data import dummy_4d
DEBUG = True  # Set to True to save 4D nii image
seg_4D = dummy_4d(vol_num=10, size_arr=(64, 64, 20), pixdim=(1,1,1),radius_RL=13.0,
 radius_AP=5.0, shape='ellipse', interleaved=True, debug=DEBUG)

[valosekj]

This factor could probably take into account breath frequency and repetition time in some way.

[jcohenadad]

it could, but i would not worry too much (as TR can also vary by several seconds)

[jcohenadad]

👍

[jcohenadad]

code duplication with lines 138-142. you could create a nested function

[valosekj]

code slightly differs in second parentheses. Isn't current workaround better for clarity than nested function?

[valosekj]

I generated some dummy 4D data with proposed dummy_segmentation_4d using following commands:

from spinalcordtoolbox.testing.create_test_data import dummy_segmentation_4d
DEBUG = True  # Set to True to save 4D nii image
seg_4D = dummy_segmentation_4d(vol_num=10, size_arr=(64, 64, 49), pixdim=(1,1,1),radius_RL=13.0,
 radius_AP=5.0, shape='ellipse', interleaved=True, debug=DEBUG)

Whole 4D data looks:

Slice-wise shift is noticeable mainly at beginning and end of SI axis. This corresponds with polynomial function fit visualization for one volume:

So, my question is - is this fit suitable? Or should it be modified to better fit middle points/slices?

I also tried to run sct_dmri_moco with interleaved option (#2664) on this dummy data. Output image is same as input, so I have to do some investigation why.

[jcohenadad]

awesome work @valosekj 🚀 yes, i think these simulated motions are fairly realistic (in terms of amplitude and location). In any case, this gives us a solid basis for improving the realism in the future (could be dedicated to another PR).