Interface method for the current registration framework using the time varying velocity field transform. More...

#include <itkTimeVaryingVelocityFieldImageRegistrationMethodv4.h>

Inheritance diagram for itk::TimeVaryingVelocityFieldImageRegistrationMethodv4< TFixedImage, TMovingImage, TTransform >:

Collaboration diagram for itk::TimeVaryingVelocityFieldImageRegistrationMethodv4< TFixedImage, TMovingImage, TTransform >:

List of all members.

Public Types
typedef CompositeTransform < RealType, ImageDimension >	CompositeTransformType
typedef SmartPointer< const Self >	ConstPointer
typedef TransformType::DerivativeType	DerivativeType
typedef TransformType::DerivativeValueType	DerivativeValueType
typedef DisplacementFieldType::Pointer	DisplacementFieldPointer
typedef TransformType::DisplacementFieldType	DisplacementFieldType
typedef TimeVaryingVelocityFieldType::PixelType	DisplacementVectorType
typedef FixedImageType::Pointer	FixedImagePointer
typedef TFixedImage	FixedImageType
typedef MetricType::Pointer	MetricPointer
typedef Superclass::MetricType	MetricType
typedef MovingImageType::Pointer	MovingImagePointer
typedef TMovingImage	MovingImageType
typedef Array< SizeValueType >	NumberOfIterationsArrayType
typedef SmartPointer< Self >	Pointer
typedef TransformType::ScalarType	RealType
typedef TimeVaryingVelocityFieldImageRegistrationMethodv4	Self
typedef ImageRegistrationMethodv4 < TFixedImage, TMovingImage, TTransform >	Superclass
typedef TimeVaryingVelocityFieldType::Pointer	TimeVaryingVelocityFieldPointer
typedef TransformType::TimeVaryingVelocityFieldType	TimeVaryingVelocityFieldType
typedef TransformOutputType::Pointer	TransformOutputPointer
typedef Superclass::TransformOutputType	TransformOutputType
typedef TransformType::Pointer	TransformPointer
typedef TTransform	TransformType
typedef MetricType::VirtualImageType	VirtualImageType
Public Member Functions
virtual ::itk::LightObject::Pointer	CreateAnother (void) const
virtual const char *	GetNameOfClass () const

virtual void	SetLearningRate (RealType _arg)
virtual RealType	GetLearningRate () const

virtual void	SetNumberOfIntegrationStepsPerTimeIndex (SizeValueType _arg)
virtual SizeValueType	GetNumberOfIntegrationStepsPerTimeIndex () const

virtual void	SetNumberOfIterationsPerLevel (NumberOfIterationsArrayType _arg)
virtual NumberOfIterationsArrayType	GetNumberOfIterationsPerLevel () const

virtual void	SetConvergenceThreshold (RealType _arg)
virtual RealType	GetConvergenceThreshold () const
Static Public Member Functions
static Pointer	New ()
Static Public Attributes
static const unsigned int	ImageDimension = TFixedImage::ImageDimension
Protected Member Functions
virtual void	GenerateData ()
virtual void	PrintSelf (std::ostream &os, Indent indent) const
virtual void	StartOptimization ()
	TimeVaryingVelocityFieldImageRegistrationMethodv4 ()
virtual	~TimeVaryingVelocityFieldImageRegistrationMethodv4 ()
Private Member Functions
void	operator= (const Self &)
	TimeVaryingVelocityFieldImageRegistrationMethodv4 (const Self &)
Private Attributes
RealType	m_ConvergenceThreshold
RealType	m_LearningRate
SizeValueType	m_NumberOfIntegrationStepsPerTimeIndex
NumberOfIterationsArrayType	m_NumberOfIterationsPerLevel

Detailed Description

template<typename TFixedImage, typename TMovingImage, typename TTransform = GaussianSmoothingOnUpdateTimeVaryingVelocityFieldTransform<double, GetImageDimension<TFixedImage>::ImageDimension>>
class itk::TimeVaryingVelocityFieldImageRegistrationMethodv4< TFixedImage, TMovingImage, TTransform >

Interface method for the current registration framework using the time varying velocity field transform.

Output: The output is the updated transform which has been added to the composite transform.

This derived class from the ImageRegistrationMethodv4 class is specialized to handle time-varying velocity field transforms of which there are currently 2:

TimeVaryingDisplacementFieldTransform
GaussianSmoothingOnUpdateTimeVaryingDisplacementFieldTransform

The latter is derived from the former and performs an optional spatial and temporal smoothing on the update and total velocity fields. Integration of the velocity field is performed using 4th order Runge Kutta and is performed using the class itkTimeVaryingVelocityFieldIntegrationImageFilter.

Optimization occurs in an iterative fashion where for each sample time point, t, in the velocity field, we integrate the velocity field in the range [0, t] to yield the displacement field which warps fixed image to time point t. Simultaneously, we integrate the velocity field in the range [t, 1] to yield the displacement field transform which warps the moving image to time point t. The metric derivative for each time point of the velocity field calculated in this way produces the normalized update field (or gradient) of the velocity field to be added to the total field at each iteration after being multiplied by the learning rate and optionally smoothed. Mathematically, this can be described as

$v_{total} = G_1( v_{total} + \lambda * G_2( v_{update} ) )$

where

$ G_1 = $ gaussian smoothing on the total field $ G_2 = $ gaussian smoothing on the update field $\lambda =$ learning rate $v_{update} =$ the normalized velocity field where we normalize the velocity field at each time point separately by the max norm of the field at that time point. This is done due to a weakly necessary (but not sufficient) condition being that the velocity field have a constant norm for all time points.