def _run_interface(self, runtime):
in_file = self.inputs.in_file
uneq_img = nb.load(in_file)
uneq_data = uneq_img.get_data()
mask = nb.load(self.inputs.mask_file)
bool_mask = mask.get_data() > 0
data_voxels = uneq_data[bool_mask]
# Do a clip on 2 to 98th percentile
bottom_2, top_98 = np.percentile(data_voxels,
np.array([1, 99]),
axis=None)
clipped_b0 = np.clip(data_voxels, 0, top_98)
eq_data = histeq(clipped_b0, num_bins=512)
output = np.zeros_like(mask.get_data())
output[bool_mask] = eq_data
eq_img = nb.Nifti1Image(output, uneq_img.affine, uneq_img.header)
self._results['out_file'] = fname_presuffix(self.inputs.in_file,
suffix='_equalized',
newpath=runtime.cwd)
eq_img.to_filename(self._results['out_file'])
return runtime
#############################################################
b0_mask, mask = median_otsu(data, , , , , ) # Creating the binary mask and the mask
mask_img = nib.Nifti1Image(mask.astype(np.float32), img.affine) # Number of bits used, in this case 32, it is saved as float32
b0_img = nib.Nifti1Image(b0_mask.astype(np.float32), img.affine) # Number of bits used, in this case 32, it is saved as float32
nib.save(mask_img, subject + '_binary_mask.nii.gz') # Just for necessary
nib.save(b0_img, subject + '_mask.nii.gz') # Just for necessary
src = join(home, '.dipy', subject + '_binary_mask.nii.gz') # Where the source file is
dst = join(dname, subject + "_binary_mask.nii.gz") # The new destination for the file
shutil.move(src, dst) # Moving the file to the right directory
src = join(home, '.dipy', subject + '_mask.nii.gz') # Where the source file is
dst = join(dname, subject + "_mask.nii.gz") # The new destination for the file
shutil.move(src, dst) # Moving the file to the right directory
#############################################################
# Plotting #
#############################################################
sli = data.shape[2] // 2
plt.figure('Brain segmentation')
plt.subplot(1, 2, 1).set_axis_off()
plt.imshow(histeq(data[:, :, sli].astype('float')).T,
cmap='gray', origin='lower')
plt.subplot(1, 2, 2).set_axis_off()
plt.imshow(histeq(b0_mask[:, :, sli].astype('float')).T,
cmap='gray', origin='lower')
b0_mask_crop, mask_crop = median_otsu(data, 4, 4, autocrop=True) # Crop the outputs to remove the largest possible number of background voxels
# May save files once decided
counter+=1 # An additional subject is completed
def dwi_to_mask(data,
subject,
affine,
outpath,
masking='median',
makefig=False,
vol_idx=None,
median_radius=5,
numpass=6,
dilate=2,
forcestart=False,
header=None,
verbose=False):
data = np.squeeze(data)
binarymask_path = os.path.join(outpath,
subject + '_dwi_binary_mask.nii.gz')
maskeddwi_path = os.path.join(outpath, subject + '_dwi_mask.nii.gz')
if not os.path.exists(binarymask_path) and not os.path.exists(
maskeddwi_path) and not forcestart:
if masking == 'median':
b0_mask, mask = median_otsu(data,
vol_idx=vol_idx,
median_radius=median_radius,
numpass=numpass,
dilate=dilate)
if masking == 'extract':
if np.size(np.shape(data)) == 3:
mask = data > 0
if np.size(np.shape(data)) == 4:
mask = data[:, :, :, 0] > 0
mask = mask.astype(np.float32)
b0_mask = applymask_array(data, mask)
if verbose:
txt = f"Creating binarymask at {binarymask_path} and masked data at {maskeddwi_path}"
print(txt)
if header is None:
save_nifti(binarymask_path, mask, affine)
save_nifti(maskeddwi_path, b0_mask.astype(np.float32), affine)
else:
binarymask_nii = nib.Nifti1Image(mask, affine, header)
nib.save(binarymask_nii, binarymask_path)
maskeddwi_nii = nib.Nifti1Image(b0_mask, affine, header)
nib.save(maskeddwi_nii, maskeddwi_path)
else:
mask = load_nifti(binarymask)
mask = mask[0]
b0_mask = load_nifti(maskeddwi)
b0_mask = b0_mask[0]
if makefig:
sli = data.shape[2] // 2
if len(b0_mask.shape) == 4:
b0_mask_2 = b0_mask[:, :, :, 0]
else:
b0_mask_2 = b0_mask
if len(data.shape) == 4:
data = data[:, :, :, 0]
plt.figure('Brain segmentation')
plt.subplot(1, 2, 1).set_axis_off()
plt.imshow(histeq(data[:, :, sli].astype('float')).T,
cmap='gray',
origin='lower')
plt.subplot(1, 2, 2).set_axis_off()
plt.imshow(histeq(b0_mask_2[:, :, sli].astype('float')).T,
cmap='gray',
origin='lower')
plt.savefig(outpath + 'median_otsu.png')
return (mask.astype(np.float32), b0_mask.astype(np.float32))
fname = 'se_1.5t'
nib.save(mask_img, fname + '_binary_mask.nii.gz')
nib.save(b0_img, fname + '_mask.nii.gz')
"""
Quick view of the results middle slice using matplotlib.
"""
import matplotlib.pyplot as plt
from dipy.core.histeq import histeq
sli = data.shape[2] / 2
plt.figure('Brain segmentation')
plt.subplot(1, 2, 1).set_axis_off()
plt.imshow(histeq(data[:, :, sli].astype('float')).T,
cmap='gray', origin='lower')
plt.subplot(1, 2, 2).set_axis_off()
plt.imshow(histeq(b0_mask[:, :, sli].astype('float')).T,
cmap='gray', origin='lower')
plt.savefig('median_otsu.png')
"""
.. figure:: median_otsu.png
:align: center
**An application of median_otsu for brain segmentation**.
``median_otsu`` can also automatically crop the outputs to remove the largest
fname = 'se_1.5t'
save_nifti(fname + '_binary_mask.nii.gz', mask.astype(np.float32), affine)
save_nifti(fname + '_mask.nii.gz', b0_mask.astype(np.float32), affine)
"""
Quick view of the results middle slice using ``matplotlib``.
"""
import matplotlib.pyplot as plt
from dipy.core.histeq import histeq
sli = data.shape[2] // 2
plt.figure('Brain segmentation')
plt.subplot(1, 2, 1).set_axis_off()
plt.imshow(
histeq(cp.asnumpy(data[:, :, sli]).astype('float')).T,
cmap='gray',
origin='lower',
)
plt.subplot(1, 2, 2).set_axis_off()
plt.imshow(histeq(b0_mask[:, :, sli].astype('float')).T,
cmap='gray',
origin='lower')
plt.savefig('median_otsu.png')
"""
.. figure:: median_otsu.png
:align: center
An application of median_otsu for brain segmentation.
