tessl/pypi-itk
Insight Toolkit for N-dimensional image processing, segmentation, and registration in medical and scientific applications
—
Pending
registration-framework.mddocs/reference/registration/
Registration Framework
Image registration frameworks for aligning images using transforms, metrics, and optimizers. Includes both classic and modern (v4) registration methods.
Capabilities
Registration Method v4 (Modern Framework)
class ImageRegistrationMethodv4:
"""
Modern registration framework (v4) supporting multi-resolution and various transform types.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
- output_transform_type: Output transform type (optional)
"""
def SetFixedImage(self, image):
"""
Set the fixed (reference) image.
Parameters:
- image: itk.Image - Fixed image
"""
def SetMovingImage(self, image):
"""
Set the moving (to be aligned) image.
Parameters:
- image: itk.Image - Moving image
"""
def SetMetric(self, metric):
"""
Set the similarity metric.
Parameters:
- metric: ImageToImageMetricv4 - Metric for comparing images
"""
def SetOptimizer(self, optimizer):
"""
Set the optimizer.
Parameters:
- optimizer: ObjectToObjectOptimizerBase - Optimizer
"""
def SetInitialTransform(self, transform):
"""
Set the initial transform.
Parameters:
- transform: itk.Transform - Initial transformation
"""
def SetMovingInitialTransform(self, transform):
"""Set initial transform for moving image."""
def SetFixedInitialTransform(self, transform):
"""Set initial transform for fixed image."""
def SetNumberOfLevels(self, levels):
"""
Set the number of multi-resolution levels.
Parameters:
- levels: int - Number of levels
"""
def SetSmoothingSigmasPerLevel(self, sigmas):
"""
Set Gaussian smoothing sigmas for each level.
Parameters:
- sigmas: list of float - Sigma for each level
"""
def SetShrinkFactorsPerLevel(self, factors):
"""
Set shrink factors for each level.
Parameters:
- factors: list of int - Shrink factors for each level
"""
def Update(self):
"""Execute registration."""
def GetTransform(self):
"""
Get the resulting transform.
Returns:
itk.Transform - Optimized transform
"""
def GetMetricValue(self):
"""Get the final metric value."""
def GetOptimizerStopConditionDescription(self):
"""Get description of why optimizer stopped."""Registration Metrics v4
class MattesMutualInformationImageToImageMetricv4:
"""
Mattes mutual information metric for registration.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
"""
def SetNumberOfHistogramBins(self, bins):
"""
Set the number of histogram bins.
Parameters:
- bins: int - Number of bins (typically 50-200)
"""
def SetUseMovingImageGradientFilter(self, use):
"""Enable precomputed moving image gradient."""
def SetUseFixedImageGradientFilter(self, use):
"""Enable precomputed fixed image gradient."""
class MeanSquaresImageToImageMetricv4:
"""
Mean squares metric for registration.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
"""
class CorrelationImageToImageMetricv4:
"""
Correlation metric for registration.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
"""
class ANTSNeighborhoodCorrelationImageToImageMetricv4:
"""
ANTs neighborhood correlation metric.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
"""
def SetRadius(self, radius):
"""
Set the neighborhood radius.
Parameters:
- radius: int or itk.Size - Radius for neighborhood
"""
class DemonsImageToImageMetricv4:
"""
Demons metric for deformable registration.
Template parameters:
- fixed_image_type: Fixed image type
- moving_image_type: Moving image type
"""Optimizers v4
class GradientDescentOptimizerv4:
"""
Gradient descent optimizer for registration.
Template parameter:
- scalar_type: Scalar type (typically double)
"""
def SetLearningRate(self, learning_rate):
"""
Set the learning rate.
Parameters:
- learning_rate: float - Step size for gradient descent
"""
def SetNumberOfIterations(self, iterations):
"""
Set the number of iterations.
Parameters:
- iterations: int - Maximum number of iterations
"""
def SetMinimumConvergenceValue(self, value):
"""Set minimum convergence threshold."""
def SetConvergenceWindowSize(self, size):
"""Set convergence window size."""
class RegularStepGradientDescentOptimizerv4:
"""
Regular step gradient descent optimizer.
Template parameter:
- scalar_type: Scalar type (typically double)
"""
def SetLearningRate(self, learning_rate):
"""Set initial learning rate."""
def SetMinimumStepLength(self, length):
"""Set minimum step length."""
def SetNumberOfIterations(self, iterations):
"""Set maximum number of iterations."""
def SetRelaxationFactor(self, factor):
"""Set relaxation factor for step length reduction."""
class ConjugateGradientLineSearchOptimizerv4:
"""
Conjugate gradient line search optimizer.
Template parameter:
- scalar_type: Scalar type (typically double)
"""
def SetLearningRate(self, learning_rate):
"""Set learning rate."""
def SetNumberOfIterations(self, iterations):
"""Set maximum number of iterations."""
class LBFGSOptimizerv4:
"""
Limited-memory BFGS optimizer.
Template parameter:
- scalar_type: Scalar type (typically double)
"""
def SetNumberOfIterations(self, iterations):
"""Set maximum number of iterations."""
def SetMaximumNumberOfFunctionEvaluations(self, evaluations):
"""Set maximum function evaluations."""
def SetGradientConvergenceTolerance(self, tolerance):
"""Set gradient convergence tolerance."""Usage Examples
Basic Registration
import itk
# Load images
fixed = itk.imread('fixed.png', itk.F)
moving = itk.imread('moving.png', itk.F)
# Set up registration
ImageType = itk.Image[itk.F, 2]
TransformType = itk.TranslationTransform[itk.D, 2]
registration = itk.ImageRegistrationMethodv4[ImageType, ImageType].New()
registration.SetFixedImage(fixed)
registration.SetMovingImage(moving)
# Create and set initial transform
initial_transform = TransformType.New()
registration.SetInitialTransform(initial_transform)
# Set metric
metric = itk.MattesMutualInformationImageToImageMetricv4[ImageType, ImageType].New()
metric.SetNumberOfHistogramBins(50)
registration.SetMetric(metric)
# Set optimizer
optimizer = itk.GradientDescentOptimizerv4Template[itk.D].New()
optimizer.SetLearningRate(1.0)
optimizer.SetNumberOfIterations(100)
registration.SetOptimizer(optimizer)
# Run registration
registration.Update()
# Get result
transform = registration.GetTransform()
print(f"Final metric value: {registration.GetMetricValue()}")
# Apply transform to moving image
resampled = itk.resample_image_filter(
moving,
transform=transform,
size=itk.size(fixed),
output_spacing=itk.spacing(fixed),
output_origin=itk.origin(fixed)
)
itk.imwrite(resampled, 'registered.png')Multi-Resolution Registration
import itk
fixed = itk.imread('fixed.png', itk.F)
moving = itk.imread('moving.png', itk.F)
ImageType = itk.Image[itk.F, 2]
registration = itk.ImageRegistrationMethodv4[ImageType, ImageType].New()
# Set images
registration.SetFixedImage(fixed)
registration.SetMovingImage(moving)
# Set multi-resolution parameters
registration.SetNumberOfLevels(3)
registration.SetSmoothingSigmasPerLevel([2.0, 1.0, 0.0])
registration.SetShrinkFactorsPerLevel([4, 2, 1])
# Create transform
transform = itk.AffineTransform[itk.D, 2].New()
registration.SetInitialTransform(transform)
# Set metric
metric = itk.MattesMutualInformationImageToImageMetricv4[ImageType, ImageType].New()
metric.SetNumberOfHistogramBins(50)
registration.SetMetric(metric)
# Set optimizer
optimizer = itk.RegularStepGradientDescentOptimizerv4Template[itk.D].New()
optimizer.SetLearningRate(1.0)
optimizer.SetMinimumStepLength(0.001)
optimizer.SetNumberOfIterations(200)
registration.SetOptimizer(optimizer)
# Run
registration.Update()
final_transform = registration.GetTransform()3D Rigid Registration
import itk
fixed = itk.imread('fixed.nii.gz', itk.F)
moving = itk.imread('moving.nii.gz', itk.F)
ImageType = itk.Image[itk.F, 3]
TransformType = itk.VersorRigid3DTransform[itk.D]
registration = itk.ImageRegistrationMethodv4[ImageType, ImageType].New()
registration.SetFixedImage(fixed)
registration.SetMovingImage(moving)
# Initialize transform with image centers
transform = TransformType.New()
initializer = itk.CenteredTransformInitializer[
TransformType,
ImageType,
ImageType
].New()
initializer.SetTransform(transform)
initializer.SetFixedImage(fixed)
initializer.SetMovingImage(moving)
initializer.GeometryOn()
initializer.InitializeTransform()
registration.SetInitialTransform(transform)
# Set metric
metric = itk.MattesMutualInformationImageToImageMetricv4[ImageType, ImageType].New()
metric.SetNumberOfHistogramBins(50)
registration.SetMetric(metric)
# Set optimizer
optimizer = itk.GradientDescentOptimizerv4Template[itk.D].New()
optimizer.SetLearningRate(0.5)
optimizer.SetNumberOfIterations(100)
registration.SetOptimizer(optimizer)
# Multi-resolution
registration.SetNumberOfLevels(3)
registration.SetSmoothingSigmasPerLevel([2.0, 1.0, 0.0])
registration.SetShrinkFactorsPerLevel([4, 2, 1])
# Run
registration.Update()
final_transform = registration.GetTransform()
# Apply transform
resampled = itk.resample_image_filter(
moving,
transform=final_transform,
size=itk.size(fixed),
output_spacing=itk.spacing(fixed),
output_origin=itk.origin(fixed),
output_direction=fixed.GetDirection()
)
itk.imwrite(resampled, 'registered_3d.nii.gz')Install with Tessl CLI
npx tessl i tessl/pypi-itk@5.4.1docs
guides
reference