def test_affine(self):
# Test affine initialization
affine_transform = pydeform.AffineTransform(
np.array((
(2, 0, 0),
(0, 3, 0),
(0, 0, 4)
)),
np.array((10, 10, 10))
)
# Do a registration pass without actual iterations to see if affine transform is
# applied to the resulting displacement field
settings = {
'max_iteration_count': 0
}
fixed = pydeform.Volume(np.zeros((10,10,10), dtype=np.float32))
moving = pydeform.Volume(np.zeros((10,10,10), dtype=np.float32))
df = pydeform.register(
fixed,
moving,
settings=settings,
affine_transform=affine_transform
)
df = np.array(df, copy=False)
# Ax + b -> A(1, 1, 1) + b -> (2, 3, 4) + (10, 10, 10) -> (12, 13, 14)
# u(x) = Ax + b - x
self.assertEqual(df[1,1,1,0], 11)
self.assertEqual(df[1,1,1,1], 12)
self.assertEqual(df[1,1,1,2], 13)
def build_mask(volume):
""" Builds a fuzzy mask for the given volume """
data = np.array(volume, copy=False)
mask = (data > 0).astype(np.float32)
mask = pydeform.Volume(mask)
mask.copy_meta_from(volume)
return mask
def _convert_image(sitk_image):
""" Converts a SimpleITK.Image to a pydeform.Volume
Return:
pydeform.Volume if sitk_image is valid, otherwise None
"""
if sitk_image is None:
return None
return pydeform.Volume(sitk.GetArrayViewFromImage(sitk_image),
sitk_image.GetOrigin(), sitk_image.GetSpacing(),
np.array(sitk_image.GetDirection()).reshape((3, 3)))
def build_regularization_map(volume, threshold, rw0, rw1):
""" Builds a regularization map given a volume. Voxels with intensities above
the given threshold have their regularization set to rw1, while all the
other voxels have regularization weight rw0.
"""
data = np.array(volume, copy=False)
regmap = np.zeros(data.shape, dtype=np.float32)
regmap = (rw0 * (data < threshold) + rw1 * (data >= threshold)).astype(
np.float32)
regmap = pydeform.Volume(regmap)
regmap.copy_meta_from(volume)
return regmap
def normalize_volume(vol):
""" Returns a new normalized copy of `vol` """
assert isinstance(vol, pydeform.Volume)
# Create a new array with a copy of the volume data
data = np.array(vol)
# Create a new normalized array
data = data / np.max(data)
# Create a new volume
out = pydeform.Volume(data)
# Copy original meta data (origin, spacing, direction)
out.copy_meta_from(vol)
return out
def downsample(vol, factor):
""" Rough downsampling of the volume by an integer factor """
assert isinstance(factor, int)
# Get the data
data = np.array(vol, copy=False)
# Create a downsampled copy of the data
data = data[::factor, ::factor, ::factor]
# Create new volume
out = pydeform.Volume(data)
# Set metadata
out.origin = vol.origin
out.spacing = np.array(vol.spacing) * factor
out.direction = vol.direction
return out