The distortion correction MI Computer is used to compute the mutual information and gradient of mutual information between two images using nonrigid, unidirectional, B-spline transform. In this variant the moving image should be a probability map. More...

#include <registration.h>

Public Member Functions
	ProbDistCorrInfoComp (bool mindiff)
	Constructor for the distortion correction class. More...

int	valueGrad (const Eigen::VectorXd &params, double &val, Eigen::VectorXd &grad)
	Computes the gradient and value of the correlation. More...

int	grad (const Eigen::VectorXd &params, Eigen::VectorXd &grad)
	Computes the gradient of the correlation. Note that this function just calls valueGrad because computing the additional values are trivial. More...

int	value (const Eigen::VectorXd &params, double &val)
	Computes the correlation. More...

size_t	nparam ()
	Returns the number of parameters (knots) More...

void	initialize (ptr< const MRImage > fixed, ptr< const MRImage > moving, size_t m_krad, size_t nbins, double space, int dir)
	Initializes the computer. This is much simipler to keep track of properly than several initialize functions that could be called in any order (eventually I'll probably move all the Comp classes to this model). More...

ptr< const MRImage >	getFixed ()
	Return the current fixed image. More...

ptr< const MRImage >	getMoving ()
	Return the current moving image. More...

ptr< MRImage >	getDeform ()
	Returns the current deformation. More...

Public Attributes
bool	m_compdiff
	Compute the difference of images (negate MI and NMI) More...

double	m_tps_reg
	Thin-plate spline regularization weight. More...

double	m_jac_reg
	Jacobian regularization weight. More...

Metric	m_metric
	Metric to use. More...

Detailed Description

The distortion correction MI Computer is used to compute the mutual information and gradient of mutual information between two images using nonrigid, unidirectional, B-spline transform. In this variant the moving image should be a probability map.

Note, if you want to register you should set the m_mindiff variable, so that the negative of mutual information will be computed. Eventually this functional will be somewhat generalized for all information-based metrics.

Definition at line 330 of file registration.h.

Constructor & Destructor Documentation

npl::ProbDistCorrInfoComp::ProbDistCorrInfoComp ( bool mindiff )

Constructor for the distortion correction class.

Parameters

mindiff Minimize difference? Set to true if you are using this for registration

Member Function Documentation

ptr<MRImage> npl::ProbDistCorrInfoComp::getDeform ( )

inline

Returns the current deformation.

Returns

Definition at line 443 of file registration.h.

ptr<const MRImage> npl::ProbDistCorrInfoComp::getFixed ( )

inline

Return the current fixed image.

Definition at line 427 of file registration.h.

ptr<const MRImage> npl::ProbDistCorrInfoComp::getMoving ( )

inline

Return the current moving image.

Note that modification of the moving image outside this class without re-calling setMoving is undefined and will result in an out-of-date moving image derivative. THIS WILL BREAK GRADIENT CALCULATIONS.

Definition at line 436 of file registration.h.

int npl::ProbDistCorrInfoComp::grad	(	const Eigen::VectorXd &	params,
		Eigen::VectorXd &	grad
	)

Computes the gradient of the correlation. Note that this function just calls valueGrad because computing the additional values are trivial.

Parameters

params	Paramters (Rx, Ry, Rz, Sx, Sy, Sz).
grad	Gradient at the given rotation

Returns: 0 if successful

void npl::ProbDistCorrInfoComp::initialize	(	ptr< const MRImage >	fixed,
		ptr< const MRImage >	moving,
		size_t	m_krad,
		size_t	nbins,
		double	space,
		int	dir
	)

Initializes the computer. This is much simipler to keep track of properly than several initialize functions that could be called in any order (eventually I'll probably move all the Comp classes to this model).

Note that modification of the moving image outside this class without re-calling setMoving is undefined and will result in an out-of-date moving image derivative. THIS WILL BREAK GRADIENT CALCULATIONS.

Parameters

fixed	New fixed image
moving	Input moving image, should be 4D with each volume containing a probability map (not modified)
m_krad	radius of kernel during histogram estimation
nbins	number of bins to estimate histograms with
dir	Direction of distortion (the dimension, must be >= 0)
space	Spacing between knots, in physical coordinates

size_t npl::ProbDistCorrInfoComp::nparam ( )

inline

Returns the number of parameters (knots)

Returns: Number of parameters that are being estimated;

Definition at line 401 of file registration.h.

int npl::ProbDistCorrInfoComp::value	(	const Eigen::VectorXd &	params,
		double &	val
	)

Computes the correlation.

Parameters

params	Paramters (Rx, Ry, Rz, Sx, Sy, Sz).
val	Value at the given rotation

Returns: 0 if successful

int npl::ProbDistCorrInfoComp::valueGrad	(	const Eigen::VectorXd &	params,
		double &	val,
		Eigen::VectorXd &	grad
	)

Computes the gradient and value of the correlation.

Parameters

params	Paramters (Rx, Ry, Rz, Sx, Sy, Sz).
val	Value at the given rotation
grad	Gradient at the given rotation

Returns: 0 if successful

Member Data Documentation

bool npl::ProbDistCorrInfoComp::m_compdiff

Compute the difference of images (negate MI and NMI)

Definition at line 379 of file registration.h.

double npl::ProbDistCorrInfoComp::m_jac_reg

Jacobian regularization weight.

Definition at line 389 of file registration.h.

Metric npl::ProbDistCorrInfoComp::m_metric

Metric to use.

Definition at line 394 of file registration.h.

double npl::ProbDistCorrInfoComp::m_tps_reg

Thin-plate spline regularization weight.

Definition at line 384 of file registration.h.

The documentation for this class was generated from the following file:

/home/micahc/coding/npl/lib/registration.h

Public Member Functions

Public Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation