Various improvements on the QC report and resampling module

[jcohenadad]

Various improvements, including:

• Moved spinalcordtoolbox/resample/nipy_resample.py --> spinalcordtoolbox/resample.py for simplicity (folder was not needed)
• Made it possible in sct_register_multimodal to have flag dseg without flag iseg (for QC).
• Added QC feature for sct_register_multimodal. Fixes add QC feature for sct_register_multimodal #1778
• QC folder name now includes milliseconds to avoid overwritting (happens when running with parallel processing)
• Now retrieve dataset and subject, according to BIDS convention, and display both in the QC report. Fixes Find another way to parse subject name #2091
• Increased kernel size for CLAHE method (was too aggressive). Also implemented contrast_stretching histogram equalization method.
• Fixed: QC generates blank images with sct_register_to_template. Fixes QC generates blank images with sct_register_to_template #2014
• QC report now compatible with uncompressed NIFTI segmentation files. Fixes QC report doesn't like uncompressed nii segmentation files #2148.
• Centralized generate_qc() within spinalcordtoolbox/reports/qc.py
• Added APRL in first image
• Use matplotlib OOP API instead of pyplot and select the backend by hand to prevent crash on headless machines. Fixes Issue while generating QC #2138

Another (unrelated) fix:

• If C2C3 automatic detection fails and is far from the cord, the straightened label will disappear from the FOV, yielding the issue described in ValueError: Array has no mass #2165. In this PR we now raise this error in case the label file is empty, suggesting the user to run the detection manually.

[jcohenadad]

@zougloub I would like to pass the new stretch_contrast_method as input param from the SCT function calling generate_qc() (e.g., here). The problem is: I don't want to add a parameter to generate_qc() and then to add_entry. I'm sure there is a more elegant way of doing this (i.e. passing the parameter once instead of three times). Any tips?

[zougloub]

@jcohenadad we can use keyword arguments (**kw) to have less code written, but it's more painful for the users as the API is less explicit (eg. IDE auto-completion doesn't work, documentation is less easy to access).

def a(**kw):
  """
  :param kw: see b()
  """
  ...
  b(**kw)
  ...

def b(**kw):
  """
  :param kw: see c()
  """
  ...
  c(**kw)
  ...

def c(p0=1, p1=2, p2=3):
  """
  :param p0: ...
  """

[jcohenadad]

I'm having issues with matplotlib's automatic adjustment of figure size upon saving. Namely, the code below (in reports/qc/_save()) does not crop identically across input examples, depending on the size:

        plt.savefig(img_path,
                    format=format,
                    bbox_inches=bbox_inches,
                    pad_inches=pad_inches,
                    transparent=True,
                    dpi=dpi)

Resulting in issues like this where the image and the segmentation is not cropped the same way (highly problematic...):

@zougloub do you have any idea how to make sure the saved figure will have a fixed number of pixels? I know matplotlib works with dpi (not fixed pix size), but i'm wondering if there is a way to ensure that, by working on the plt object directly.

[jcohenadad]

@zougloub @perone Regarding the "WARNING" (#2157 (comment)) about manually deleting the resample/ folder after a git pull (because git keeps track of files, not folders), I'm afraid this will cause a lot of trouble in our users. Can you recommend something to deal with this problem? E.g. added some code in the installation to delete empty folders, etc.

[zougloub]

I have no clean solution, just workarounds (eg. renaming the new module resampling or leaving stuff in the __init__.py...), and IMO doing something in the installer is worse.

[charleygros]

* Moved `spinalcordtoolbox/resample/nipy_resample.py` --> `spinalcordtoolbox/resample.py` for simplicity (folder was not needed)

--> Tested with sct_resample and sct_deepseg_sc, seems all good to me.

* QC folder name now includes milliseconds to avoid overwritting (happens when running with parallel processing)

--> Tested, seems good to me.

* Now retrieve `dataset` and `subject`, according to BIDS convention, and display both in the QC report. Fixes #2091

--> Tested on a dataset with BIDS convention, works!!

* QC report now compatible with uncompressed NIFTI segmentation files. Fixes #2148.

--> Tested, works well.

* Added APRL in first image

--> Tested, really Nice.