sct_get_centerline: Use the new viewer

[charleygros]

sct_get_centerline currently uses sct_viewer.ClickViewerPropseg() instead of the new spinalcordtoolbox.gui.* viewer.

This Pull Request aims at integrating the new viewer into sct_get_centerline.

To test it, please run:

cd example_data/t2
sct_get_centerline -i t2.nii.gz -method viewer
sct_image -i t2.nii.gz -setorient SAL -o t2SAL.nii.gz
sct_get_centerline -i t2SAL.nii.gz -method viewer

This code was in sct_deepseg_sc --> I moved it in sct_get_centerline --> that s why both sct_deepseg_sc and sct_deepseg_lesion now called functions from sct_get_centerline.

To test it, please run:

cd example_data/t2
sct_deepseg_sc -i t2.nii.gz -c t2 -centerline viewer

This PR fixes the issue #1837.

[jcohenadad]

there is a weird bug (i think related to the centerline smoothing). To reproduce:

cd sct_example_data/t1/
sct_get_centerline -i t1.nii.gz -method viewer

results (tested on ba40d2f):

[charleygros]

@jcohenadad : Thank you. I reproduced the bug, and I think it comes from this line:
https://github.com/neuropoly/spinalcordtoolbox/blob/master/scripts/sct_straighten_spinalcord.py#L159

I will investigate in this way.

UPDATE: In the commit 983a719, I reduced the nurbs_pts_number in the smoothing. It seems to fix it.

[jcohenadad]

i still find issues if the number of points is too low (switches to hanning, but we don't know because verbose is set to False). For this PR, i would suggest to only introduce the new viewer, without perturbating too much the behaviour (which requires extensive testing on larger set of images). So i would suggest the following modifs:

• set verbose to "verbose" (instead of 0) when calling smooth_centerline
• set nurbs_pts_number=3000 (as was the case before)
• use the same number of points & gap as before (was calculated based on max number of slices)
• if not too much trouble, replace _from_viewerLabels_to_centerline by the already-existing sct_process_segmentation.extract_centerline (which does the same thing)

happy to help :-)

[charleygros]

@jcohenadad: Thank you for your help.
The last commits do:

• replace _from_viewerLabels_to_centerline by the already-existing sct_process_segmentation.extract_centerline
• set verbose to "verbose" (instead of 0) when calling smooth_centerline in sct_process_segmentation.extract_centerline
• use the same number of points & gap as before (calculated based on max number of slices)

I tested it on the example_data with the default param (gap=10) and it seems to work well.
Please let me know your thoughts

[jcohenadad]

i find it a bit too dense now, how about setting a compromise with interslice_gap=20?

[charleygros]

I agree!

[jcohenadad]

great contribution! I just made a minor comment about the default value for interslice_gap. Otherwise looks OK!

[charleygros]

@jcohenadad: By increasing the default gap to 20.0mm, I observed 'not smooth centerline' again.
In the last commits, I proposed to increase the number of points used by the nurbs (3000 to 8000) --> looks much better. However it takes more time.

Another approach would be to leave gap=20 and nurbs_pts_number=3000 but to send a message to the user if: https://github.com/neuropoly/spinalcordtoolbox/blob/master/scripts/sct_straighten_spinalcord.py#L160
--> asking the user to decrease the gap used by the viewer.

[jcohenadad]

@charleygros let's go with your latest commit. We definitely don't want to have an approach where a message is sent, asking the user to make a decision (99.99% of our users don't read the terminal output).