__all__ = ['RigidRegistration']
from ...registration import interface, apply_transforms
class Registration(object):
"""
How would it work:
# Co-registration within-visit
reg = Registration('Rigid', fixed_image=t1_template,
save_dir=save_dir, save_suffix='_coreg')
for img in other_imgs:
reg.fit(img)
for img in [flair_img, t2_img]:
reg = Registration('Rigid', fixed_image=t1_template,
save_dir=save_dir, save_suffix='_coreg')
reg.fit(img)
# Co-registration across-visit
reg = Registration('Rigid', fixed_image=t1)
reg.fit(moving=t1_followup)
# now align all followups with first visit
for img in [flair_follwup, t2_followup]:
img_reg = reg.transform(img)
# conversly, align all first visits with followups
for img in [flair, t2]:
img_reg = reg.inverse_transform(img)
"""
def __init__(self, type_of_transform, fixed_image):
"""
Properties:
type_of_transform
fixed_image (template)
save_dir (where to save outputs)
save_suffix (what to append to saved outputs)
save_prefix (what to preppend to saved outputs)
"""
self.type_of_transform = type_of_transform
self.fixed_image = fixed_image
def fit(self, X, y=None):
"""
X : ANTsImage | string | list of ANTsImage types | list of strings
images to register to fixed image
y : string | list of strings
labels for images
"""
moving_images = X if isinstance(X, (list,tuple)) else [X]
moving_labels = y if y is not None else [i for i in range(len(moving_images))]
fixed_image = self.fixed_image
self.fwdtransforms_ = {}
self.invtransforms_ = {}
self.warpedmovout_ = {}
self.warpedfixout_ = {}
for moving_image, moving_label in zip(moving_images, moving_labels):
fit_result = interface.registration(fixed_image,
moving_image,
type_of_transform=self.type_of_transform,
initial_transform=None,
outprefix='',
mask=None,
grad_step=0.2,
flow_sigma=3,
total_sigma=0,
aff_metric='mattes',
aff_sampling=32,
syn_metric='mattes',
syn_sampling=32,
reg_iterations=(40,20,0),
verbose=False)
self.fwdtransforms_[moving_label] = fit_result['fwdtransforms']
self.invtransforms_[moving_label] = fit_result['invtransforms']
self.warpedmovout_[moving_label] = fit_result['warpedmovout']
self.warpedfixout_[moving_label] = fit_result['warpedfixout']
return self
def transform(self, X, y=None):
pass
class RigidRegistration(object):
"""
Rigid Registration as a Scikit-Learn compatible transform class
Example
-------
>>> import ants
>>> import ants.extra as extrants
>>> fi = ants.image_read(ants.get_data('r16'))
>>> mi = ants.image_read(ants.get_data('r64'))
>>> regtx = extrants.RigidRegistration()
>>> regtx.fit(fi, mi)
>>> mi_r = regtx.transform(mi)
>>> ants.plot(fi, mi_r.iMath_Canny(1, 2, 4).iMath('MD',1))
"""
def __init__(self, fixed_image=None):
self.type_of_transform = 'Rigid'
self.fixed_image = fixed_image
def fit(self, moving_image, fixed_image=None):
if fixed_image is None:
if self.fixed_image is None:
raise ValueError('must give fixed_image in fit() or set it in __init__')
fixed_image = self.fixed_image
fit_result = interface.registration(fixed_image,
moving_image,
type_of_transform=self.type_of_transform,
initial_transform=None,
outprefix='',
mask=None,
grad_step=0.2,
flow_sigma=3,
total_sigma=0,
aff_metric='mattes',
aff_sampling=32,
syn_metric='mattes',
syn_sampling=32,
reg_iterations=(40,20,0),
verbose=False)
self._fit_result = fit_result
self.fwdtransforms_ = fit_result['fwdtransforms']
self.invtransforms_ = fit_result['invtransforms']
self.warpedmovout_ = fit_result['warpedmovout']
self.warpedfiout_ = fit_result['warpedfixout']
def transform(self, moving_image, fixed_image=None):
result = apply_transforms(fixed=fixed_image, moving=moving_image,
transformlist=self.fwdtransforms)
return result
def inverse_transform(self, moving_image, fixed_image=None):
result = apply_transforms(fixed=fixed_image, moving=moving_image,
transformlist=self.invtransforms)
return result
