FREE::ConstrainedAccuracySOMetric< VImageDimension > Class Template Reference
[Setup optimization metric]

This class is a metric which evaluates setups with respect of there accuracy. More...

#include <freConstrainedAccuracySOMetric.h>

Inheritance diagram for FREE::ConstrainedAccuracySOMetric< VImageDimension >:

Collaboration diagram for FREE::ConstrainedAccuracySOMetric< VImageDimension >:

List of all members.

Public Types

typedef
ConstrainedAccuracySOMetric
< VImageDimension > Self

typedef
AccuracySOMetric
< VImageDimension > Superclass

typedef
itk::SmartPointer
< Self > Pointer

typedef
itk::SmartPointer
< const Self > ConstPointer

typedef
Superclass::DecomposedMeasureType DecomposedMeasureType

Public Member Functions

virtual const char * GetNameOfClass () const

virtual
::itk::LightObject::Pointer CreateAnother (void) const

virtual unsigned int GetNumberOfValues (void) const

Static Public Member Functions

static Pointer New (void)

Protected Types

typedef
SetupOptimizationMetric::ParametersType ParametersType

Protected Member Functions

ConstrainedAccuracySOMetric ()

virtual
DecomposedMeasureType ComputeDecomposedValue (const ParametersType &parameters) const

Private Member Functions

ConstrainedAccuracySOMetric (const Self &)

void operator= (const Self &)

Detailed Description

template<unsigned int VImageDimension>
class FREE::ConstrainedAccuracySOMetric< VImageDimension >

This class is a metric which evaluates setups with respect of there accuracy.

ConstrainedAccuracySOMetric is used to estimate the accuracy of a registration by two options:
1. comparing resulting deformation fields with given reference deformation fields. The reference field of every adaptation must be stored in a media container. The IDPath to this media can be set via ReferenceFieldPath parameter.
2. by using the resulting deformation field to check if reference points of the moving image match the points of the reference. The IDPath to both media can be set via ReferencePointPath and MovingPointPath.
In addition to the normal AccuracySOMetric, the basic objective function can be modified by additional constraints, which will also produce metric values and are handled as barrier functions.
Equality constraints will result 0 as long as the constraint is met. Otherwise they will result MaxConstraintPenalty.
Inequality constraints will be transformed to an inequality greater or equal zero (c(x)>= 0). The BarrierSize indicates the size of the region in front of the border of the legal value range. The barrier function is modelled by log(c(x)/BarrierSize). Example:
The given constraint is 500<= param_1. Its barriersize is 50, so the barrier region is between 500 and 550.
The constraint will be transformed in 0 <= param_1 - 500. The barrier function will be 0 for any param_1 > 550 and min(-log((param_1 - 500)/50), MaxConstraintPenalty) for any other value of param_1.
The Metric has multiple metric values as return:

absolute mean error (occurring when registrating the whole adaptation list)
mean error standard deviation
min absolute error
max absolute error
number of failed registrations (e.g. all points mapped outside the image, hence metric throws exception)
number of reference points that could not be evaluated, because they where outside the transformation field (if a reference field is used for comparison, this value is always 0)
mean duration of registration (in tenth of a second (ToS))
standard deviation of the duration (in ToS)
min duration (in ToS)
max duration (in ToS)
n constraint penalties (one value for each constrained).

Remarks:: This version of an accuracy metric is very useful if you want to regard several constraints while optimizing, but the optimizer supports no constraints by default (e.g. amoeba simplex optimizer or Powell optimizer). This metric bypasses this problem by adding the constraints as boundary functions to the cost functions as additional metric values.

Definition at line 85 of file freConstrainedAccuracySOMetric.h.