sct_deepseg_sc - Issue when input is .nii instead of .nii.gz

[charleygros]

Requirements

This issue was reported twice:

• by @stephaniealley : Issues with segmentation results #1634
• by @SaraDupont : Test DeepSeg on SCI patients from UCSF  #1659

Description of the Change

>>> from msct_image import Image
>>> s_compressed=Image('t2_seg.nii.gz')
>>> type(s_compressed.data)
<type 'numpy.ndarray'>
>>> s_not_compressed=Image('t2_seg.nii')
>>> type(s_not_compressed.data)
<class 'numpy.core.memmap.memmap'>

>>> from scipy.ndimage.measurements import center_of_mass
>>> from msct_image import Image

>>> s_not_compressed = Image('t2_seg.nii')
>>> type(s_not_compressed.data)
<class 'numpy.core.memmap.memmap'>
>>> center_of_mass(s_not_compressed.data)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/chgroc/spinalcordtoolbox/python/lib/python2.7/site-packages/scipy/ndimage/measurements.py", line 1289, in center_of_mass
    return [tuple(v) for v in numpy.array(results).T]
TypeError: 'numpy.float64' object is not iterable

>>> s_compressed = Image('t2_seg.nii.gz')
>>> type(s_compressed.data)
<type 'numpy.ndarray'>
>>> center_of_mass(s_compressed.data)
(225.70722563053852, 216.90064758009544, 7.8552317655078392)

--> Error when scipy.ndimage.measurements.center_of_mass receives a numpy.core.memmap.memmap as input.

Proposed solution:
Cast np.memmap into np.array before to apply scipy.ndimage.measurements.center_of_mass.

Is there a better approach?

---

Additional tests were done on ~30 data (different contrast, centers...etc) --> works

[zougloub]

numpy limitation (other people have seen it: https://stackoverflow.com/a/25370699 / numpy/numpy#667) and we still have it because we're bundling an archaic version of numpy.

[zougloub]

I'd put the np.array() cast right in there, so it is more obvious that we're doing it for center_of_mass().

[charleygros]

True, I am modifying it

[zougloub]

You may amend your commit and re-push your branch to have the corresponding issue ids in the commit title. But it's not bad as is.