def register2D(fixed,
moving,
numberOfBins,
sampling_percentage=0.1,
learning_rate=1.1,
min_step=0.001,
relaxation_factor=0.8):
best_resampler = reg.register(fixed, moving, numberOfBins,
sampling_percentage)
return best_resampler
def register2D(fixed,
moving,
numberOfBins,
is_best_rotation=False,
array_moving=None,
flipped=False,
sampling_percentage=0.1,
learning_rate=1.1,
min_step=0.001,
relaxation_factor=0.8):
to_flip = False
# Flip axis and choose best fit during registration
if flipped:
# Construct a 3D image
# 2 slices = original + flipped
flipped_moving = sitk.Flip(moving, (True, False))
moving_and_flipped = np.zeros(
(2, moving.GetSize()[1], moving.GetSize()[0]), dtype=np.float32)
moving_and_flipped[0, ...] = sitk.GetArrayFromImage(moving)
moving_and_flipped[1, ...] = sitk.GetArrayFromImage(flipped_moving)
array_flipped = moving_and_flipped
#determines if symmetry of image is a better match
best_resampler, index = reg.best_fit(
fixed,
array_flipped,
numberOfBins,
sampling_percentage,
find_best_rotation=is_best_rotation,
learning_rate=learning_rate,
min_step=min_step,
relaxation_factor=relaxation_factor)
if index == 1:
to_flip = True
else:
#normal registration
best_resampler = reg.register(fixed,
moving,
numberOfBins,
sampling_percentage,
find_best_rotation=is_best_rotation,
learning_rate=learning_rate,
min_step=min_step,
relaxation_factor=relaxation_factor)
return best_resampler, to_flip
def registration_number_bins(outputname, min_bins, max_bins):
"""
Registration over a range of bins
for the sampling strategy
Parameters
----------
outputname: str
output image name
min_bins: int
minimum number of bins
max_bins: int
maximum number of bins
"""
for i in range(min_bins, max_bins):
outname = os.path.splitext(outputname)[0] + "_bins" + str(i) + ".png"
try:
resampler = reg.register(fixed, moving, i)
out = resampler.Execute(moving)
out = sitk.Cast(sitk.RescaleIntensity(out), sitk.sitkUInt8)
sitk.WriteImage(out, outname)
except Exception as e:
pass
normalized_dt_reference_itk = sitk.GetImageFromArray(normalized_dt_reference)
normalized_dt_target_itk = sitk.GetImageFromArray(normalized_dt_target)
normalized_dt_reference_itk = utils.local_max_dt(reference_itk)
normalized_dt_target_itk = utils.local_max_dt(target_itk)
# fig, ax = plt.subplots(1, 2)
# ax[0].imshow(normalized_dt_reference)
# ax[1].imshow(normalized_dt_target)
# plt.show()
R = reg.register(reference_itk,
target_itk,
15,
0.1,
find_best_rotation=True,
learning_rate=0.00001,
update_DT=False,
normalize_DT=False)
out = R.Execute(dt_target)
R_update = reg.register(normalized_dt_reference_itk,
normalized_dt_target_itk,
15,
0.1,
find_best_rotation=True,
learning_rate=0.00001,
update_DT=False,
normalize_DT=True)
out_norm = R_update.Execute(normalized_dt_target_itk)
out_update = R_update.Execute(dt_target)
target = np.pad(target, 25) noise = np.random.normal(mean_noise, stddev_noise, reference.shape) # reference += noise # target += noise reference_itk = sitk.GetImageFromArray(reference) target_original_itk = sitk.GetImageFromArray(target_original) target_itk = sitk.GetImageFromArray(target) dt_reference = utils.compute_DT(reference_itk) dt_target = utils.compute_DT(target_itk) dt_target_original = utils.compute_DT(target_original_itk) R_intensity = reg.register(reference_itk, target_itk, 10, 0.001, find_best_rotation=True, use_DT=False, learning_rate=0.00000001) out_intensity = R_intensity.Execute(target_itk) R_shape = reg.register(reference_itk, target_itk, 10, 0.001, find_best_rotation=True, learning_rate=0.00000001) out_shape = R_shape.Execute(target_itk) sitk.WriteImage(sitk.Cast(out_intensity, sitk.sitkUInt8), "registered_intensity.tif") sitk.WriteImage(sitk.Cast(out_shape, sitk.sitkUInt8), "registered_shape.tif") sitk.WriteImage(dt_reference, "dt_reference.tif") sitk.WriteImage(dt_target, "dt_target.tif") sitk.WriteImage(dt_target_original, "dt_target_original.tif") sitk.WriteImage(sitk.Cast(reference_itk, sitk.sitkUInt8), "reference.tif") sitk.WriteImage(sitk.Cast(target_itk, sitk.sitkUInt8), "target.tif") sitk.WriteImage(sitk.Cast(target_original_itk, sitk.sitkUInt8), "target_original.tif")