def write_separate_labels(hard_segmentation_path):
""" Write binary arrays for each of the labels separately. """
# construct the binary arrays
hard_map = oitk.get_itk_array(hard_segmentation_path)
enhanced = (hard_map == 1).astype('uint8')
tumor_core = (np.logical_and(enhanced, hard_map == 2)).astype('uint8')
whole_tumor = (hard_map > 0).astype('uint8')
# get the paths to write the arrays to
dirname = os.path.dirname(hard_segmentation_path)
enhanced_path = os.path.join(dirname, 'enhanced.nii.gz')
tumor_core_path = os.path.join(dirname, 'tumor_core.nii.gz')
whole_tumor_path = os.path.join(dirname, 'whole_tumor.nii.gz')
# write
itk_proto = oitk.get_itk_image(hard_segmentation_path)
oitk.write_itk_image(oitk.make_itk_image(enhanced, itk_proto),
enhanced_path)
oitk.write_itk_image(oitk.make_itk_image(tumor_core, itk_proto),
tumor_core_path)
oitk.write_itk_image(oitk.make_itk_image(whole_tumor, itk_proto),
whole_tumor_path)
def get_non_uniform_supervoxels(slic_im, hard_map):
""" Get supervoxel indices of supervoxels in slic_im
which have multiple values in hard_map. """
# Quick check
slic_ints = np.unique(slic_im[hard_map != 0])
a = sitk.LabelStatisticsImageFilter()
a.Execute(oitk.make_itk_image(hard_map, verbose=False),
oitk.make_itk_image(slic_im, verbose=False))
slics_to_split = [i for i in slic_ints \
if a.GetSigma(int(i))>0]
return slics_to_split
def run(self, scan, output, mask=None):
img = oitk.get_itk_image(scan)
image = oitk.get_itk_array(img)
if mask is None:
mask = self.generateMask(image, self.thresh)
if mask is False:
mask = np.zeros(image.shape)
mask[image > 0] = 1
normalized = self.normalize(image, mask)
oitk.write_itk_image(oitk.make_itk_image(normalized, proto_image=img),
os.path.join(output,
os.path.split(scan)[1]))
def save_segm(self):
""" Save the current custom segmentation into a file."""
kwargs = {}
if self.segm_path is not None:
dirname, basename = os.path.split(self.segm_path)
if not basename.startswith('corrected_'):
basename = 'corrected_'+basename
kwargs['initialfile'] = basename
kwargs['initialdir'] = dirname
path = asksaveasfilename(title="Please select a path to save your segmentation",
filetypes=[('Image files',
('.nii',
'.mha',
'.nii.gz'))],
**kwargs)
#TODO: change here self.segm_path
if self.segm_path is not None:
old_segm = oitk.get_itk_image(self.segm_path)
image = oitk.make_itk_image(self.segm, old_segm)
oitk.write_itk_image(image, path)
def get_borders(supervoxels):
""" Get supervoxel borders as a binary array. Voxels on the edge of each
nonzero-labeled object are 1, all others are set to 0.
Parameters
----------
supervoxels : np.ndarray
integer mask of the supervoxels
Returns
-------
border_arr : np.ndarray
array of the same shape as supervoxels, with voxels on the edge of non-zero
supervoxels set to 1 and all others to zero.
"""
border_im = sitk.LabelContour(
oitk.make_itk_image(supervoxels, verbose=False))
border_arr = oitk.get_itk_array(border_im)
border_arr = border_arr > 0
return border_arr