def test_warp_image_ddf():
"""
Test warp_image_ddf by checking input/output shapes
"""
batch_size = 2
fixed_image_size = (32, 32, 16)
moving_image_size = (24, 24, 16)
channel = 6
image = tf.ones((batch_size, *moving_image_size), dtype="float32")
image_ch = tf.ones((batch_size, *moving_image_size, channel),
dtype="float32")
ddf = tf.ones((batch_size, *fixed_image_size, 3), dtype="float32")
grid_ref = tf.ones((1, *fixed_image_size, 3), dtype="float32")
# without channel, with grid_ref
got = layer_util.warp_image_ddf(image=image, ddf=ddf, grid_ref=grid_ref)
assert got.shape == (batch_size, *fixed_image_size)
# without channel, without grid_ref
got = layer_util.warp_image_ddf(image=image, ddf=ddf, grid_ref=None)
assert got.shape == (batch_size, *fixed_image_size)
# with channel, with grid_ref
got = layer_util.warp_image_ddf(image=image_ch, ddf=ddf, grid_ref=grid_ref)
assert got.shape == (batch_size, *fixed_image_size, channel)
# with channel, without grid_ref
got = layer_util.warp_image_ddf(image=image_ch, ddf=ddf, grid_ref=None)
assert got.shape == (batch_size, *fixed_image_size, channel)
# wrong image shape
wrong_image = tf.ones(moving_image_size, dtype="float32")
with pytest.raises(ValueError) as err_info:
layer_util.warp_image_ddf(image=wrong_image,
ddf=ddf,
grid_ref=grid_ref)
assert "image shape must be (batch, m_dim1, m_dim2, m_dim3)" in str(
err_info.value)
# wrong ddf shape
wrong_ddf = tf.ones((batch_size, *fixed_image_size, 2), dtype="float32")
with pytest.raises(ValueError) as err_info:
layer_util.warp_image_ddf(image=image,
ddf=wrong_ddf,
grid_ref=grid_ref)
assert "ddf shape must be (batch, f_dim1, f_dim2, f_dim3, 3)" in str(
err_info.value)
# wrong grid_ref shape
wrong_grid_ref = tf.ones((batch_size, *moving_image_size, 3),
dtype="float32")
with pytest.raises(ValueError) as err_info:
layer_util.warp_image_ddf(image=image,
ddf=ddf,
grid_ref=wrong_grid_ref)
assert "grid_ref shape must be (1, f_dim1, f_dim2, f_dim3, 3) or None" in str(
err_info.value)
def gif_warp(
img_paths, ddf_path, slice_inds=None, num_interval=100, interval=50, save_path=""
):
"""
Apply ddf to image slice/s to generate gif
:param img_paths: list or comma separated string of image paths
:param ddf_path: path to ddf to use for warping
:param slice_inds: list of slice indices to use for each image
:param num_interval: number of intervals in which to apply ddf
:param interval: time in miliseconds between frames of gif
:param save_path: path to directory where visualisation/s is/are to be saved
"""
if type(img_paths) is str:
img_paths = string_to_list(img_paths)
image = load_nifti_file(img_paths[0])
img_shape = np.shape(image)
if slice_inds is None:
slice_inds = [round(np.random.rand() * (img_shape[-1]) - 1)]
for img_path in img_paths:
for slice_ind in slice_inds:
fig = plt.figure()
ax = plt.Axes(fig, [0.0, 0.0, 1.0, 1.0])
ax.set_axis_off()
fig.add_axes(ax)
ddf_scalers = np.linspace(0, 1, num=num_interval)
frames = []
for ddf_scaler in ddf_scalers:
image = load_nifti_file(img_path)
ddf = load_nifti_file(ddf_path)
image = np.expand_dims(image, axis=0)
ddf = np.expand_dims(ddf, axis=0) * ddf_scaler
warped_image = warp_image_ddf(image=image, ddf=ddf, grid_ref=None)
warped_image = np.squeeze(warped_image.numpy())
frame = plt.imshow(
warped_image[:, :, slice_ind], aspect="auto", animated=True
)
frames.append([frame])
anim = animation.ArtistAnimation(fig, frames, interval=interval)
path_to_anim_save = os.path.join(
save_path,
os.path.split(img_path)[-1].split(".")[0]
+ "_slice_"
+ str(slice_ind)
+ ".gif",
)
anim.save(path_to_anim_save)
print("Animation saved to:", path_to_anim_save)
def call(self, inputs, **kwargs) -> tf.Tensor:
"""
:param inputs: (ddf, image)
- ddf, shape = (batch, f_dim1, f_dim2, f_dim3, 3), dtype = float32
- image, shape = (batch, m_dim1, m_dim2, m_dim3), dtype = float32
:param kwargs: additional arguments.
:return: shape = (batch, f_dim1, f_dim2, f_dim3)
"""
return layer_util.warp_image_ddf(
image=inputs[1], ddf=inputs[0], grid_ref=self.grid_ref
)
def call(self, inputs, **kwargs):
"""
wrap an image into a fixed size using ddf
same functionality as transform of neuron
https://github.com/adalca/neuron/blob/master/neuron/utils.py
vol = image
loc_shift = ddf
:param inputs: [ddf, image]
ddf.shape = (batch, f_dim1, f_dim2, f_dim3, 3)
image.shape = (batch, m_dim1, m_dim2, m_dim3)
ddf.type = float32
image.type = float32
:param kwargs:
:return: shape = (batch, f_dim1, f_dim2, f_dim3)
"""
return layer_util.warp_image_ddf(image=inputs[1],
ddf=inputs[0],
grid_ref=self.grid_ref)
def warp(image_path: str, ddf_path: str, out_path: str):
"""
:param image_path: file path of the image file
:param ddf_path: file path of the ddf file
:param out_path: file path of the output
"""
if out_path == "":
out_path = "warped.nii.gz"
logging.warning(
f"Output file path is not provided, will save output in {out_path}."
)
else:
if not (out_path.endswith(".nii") or out_path.endswith(".nii.gz")):
out_path = os.path.join(os.path.dirname(out_path), "warped.nii.gz")
logging.warning(
f"Output file path should end with .nii or .nii.gz, "
f"will save output in {out_path}.")
os.makedirs(os.path.dirname(out_path), exist_ok=True)
# load image and ddf
image = load_nifti_file(image_path)
ddf = load_nifti_file(ddf_path)
if len(image.shape) not in [3, 4]:
raise ValueError(f"image shape must be (m_dim1, m_dim2, m_dim3) "
f"or (m_dim1, m_dim2, m_dim3, ch),"
f" got {image.shape}")
if not (len(ddf.shape) == 4 and ddf.shape[-1] == 3):
raise ValueError(
f"ddf shape must be (f_dim1, f_dim2, f_dim3, 3), got {ddf.shape}")
# add batch dimension manually
image = tf.expand_dims(image, axis=0)
ddf = tf.expand_dims(ddf, axis=0)
# warp
warped_image = warp_image_ddf(image=image, ddf=ddf, grid_ref=None)
warped_image = warped_image.numpy()
warped_image = warped_image[0, ...] # removed added batch dimension
# save output
nib.save(img=nib.Nifti2Image(warped_image, affine=np.eye(4)),
filename=out_path)
def warp(image_path: str, ddf_path: str, out_path: str):
"""
:param image_path: file path of the image file
:param ddf_path: file path of the ddf file
:param out_path: file path of the output
"""
if out_path == "":
out_path = "warped.nii.gz"
logging.warning(
f"Output file path is not provided, will save output in {out_path}."
)
else:
if not (out_path.endswith(".nii") or out_path.endswith(".nii.gz")):
out_path = os.path.join(os.path.dirname(out_path), "warped.nii.gz")
logging.warning(
f"Output file path should end with .nii or .nii.gz, "
f"will save output in {out_path}."
)
os.makedirs(os.path.dirname(out_path), exist_ok=True)
# load image and ddf
image = load_nifti_file(image_path)
ddf = load_nifti_file(ddf_path)
shape_sanity_check(image=image, ddf=ddf)
# add batch dimension manually
image = tf.expand_dims(image, axis=0)
ddf = tf.expand_dims(ddf, axis=0)
# warp
warped_image = warp_image_ddf(image=image, ddf=ddf, grid_ref=None)
warped_image = warped_image.numpy()
warped_image = warped_image[0, ...] # removed added batch dimension
# save output
nib.save(img=nib.Nifti1Image(warped_image, affine=np.eye(4)), filename=out_path)