Package gov.nih.mipav.model.algorithms

Class AlgorithmELSUNCOpt3D

java.lang.Object

java.lang.Thread

gov.nih.mipav.model.algorithms.AlgorithmBase

gov.nih.mipav.model.algorithms.AlgorithmELSUNCOpt3D

All Implemented Interfaces:

ActionListener, WindowListener, Runnable, EventListener

public class AlgorithmELSUNCOpt3D
extends AlgorithmBase

Runs ELSUNC, LEVENBERG_MARQUARDT, or NL2SOL for a 3D image.

Based on ELSUNC allowed by the author with acknowledgement:

Gauss-Newton Based Algorithms For Constrained Nonlinear Least Squares Problems by Per Lindstrom and Per-Ake Wedin, Institute of Information Processing, University of Umea, S-901 87 Umea, Sweden This can be downleaded from http://www.cs.umu.se/~perl/reports/alg.ps.gz

Version:

0.1 March 23, 2012

Author:

William Gandler

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
(package private) class	AlgorithmELSUNCOpt3D.FitOAR3DELSUNCModel
(package private) class	AlgorithmELSUNCOpt3D.FitOAR3DLMModel
(package private) class	AlgorithmELSUNCOpt3D.FitOAR3DNL2solModel

Nested classes/interfaces inherited from class java.lang.Thread

Thread.Builder, Thread.State, Thread.UncaughtExceptionHandler

Field Summary

Fields

Modifier and Type	Field	Description
private AlgorithmOptimizeFunctionBase	costFunction	Cost function called to measure cost - 1D.
private final int	ELSUNC
(package private) AlgorithmELSUNCOpt3D.FitOAR3DELSUNCModel	eModel
private double	epsilon
private TransMatrixd	fromOrigin	The transformation matrix from the origin of the input image.
private double	functionAtBest	The cost of the function at the best minimum.
private double	huge
private final int	LEVENBERG_MARQUARDT
(package private) AlgorithmELSUNCOpt3D.FitOAR3DLMModel	lmModel
private int	maxIterations	The maximum number of iterations the optimization allows.
private double	minFunctionAtBest
private int	nDims	Degress of freedom.
private final int	NL2SOL
(package private) AlgorithmELSUNCOpt3D.FitOAR3DNL2solModel	nModel
private double[]	OARTolerance	Array of tolerances for each dimension.
private AlgorithmBase	parent	Parent algorithm that called this optimization.
private Vectornd[]	points	Array used to hold the initial points, final points and costs
private int	searchAlgorithm
private double[]	start	Point that was initially passed into function.
private int	status
private TransMatrixd	toOrigin	The transformation matrix to the origin of the input image.

Fields inherited from class gov.nih.mipav.model.algorithms.AlgorithmBase

destFlag, destImage, image25D, mask, maxProgressValue, minProgressValue, multiThreadingEnabled, nthreads, progress, progressModulus, progressStep, runningInSeparateThread, separable, srcImage, threadStopped

Fields inherited from class java.lang.Thread

MAX_PRIORITY, MIN_PRIORITY, NORM_PRIORITY

Constructor Summary

Constructors

Constructor	Description
AlgorithmELSUNCOpt3D(AlgorithmBase parent, WildMagic.LibFoundation.Mathematics.Vector3f com, int degreeOfFreedom, AlgorithmOptimizeFunctionBase costFunc, double[] tols, int maxIter, int searchAlgorithm)	Constructs a new algorithm with the given centers of mass (needed for setting the transformations), the given cost function (which was constructed with the proper images), the initial point we're looking at, some tolerance within that point to look for the minimum, and the maximum number of iterations.

Method Summary

Modifier and Type	Method	Description
void	adjustTranslation(TransMatrixd mat, float sample)
double[]	constructPoint(double[] defaultPoint, double[] point)	Construct a full 12-dimension transformation vector from the partial transformation vector.
TransMatrixd	convertToMatrix(double[] vector)	Convert a transformation vector to a transformation matrix.
TransMatrixd	convertToMatrix(TransMatrixd toOrigin, TransMatrixd fromOrigin, double[] vector)	Convert a 12-dimension transformation vector to a 4x4 transformation matrix.
double[]	convertToMidsagittal(double[] point)	Converts a full transformation vector into a midsagittal transformation vector.
static final double[]	copyPoint(double[] point)	Make a copy of the transformation vector
private void	dfault(int[] iv, double[] v)
void	disposeLocal()	Sets everything to null and prepares this class for destruction.
double[]	extractPoint(double[] startPoint)	Extract the partial or full transformation vector from the start transformation vector, which will be optimized.
final double	getCost(int index)	Returns the cost for the transformation vector.
double[]	getFinal(double[] point)	Accessor that returns the final point with translations, rotations, scales, and skews representing the best tranformation.
TransMatrixd	getMatrix(double[] point, float sample)
final TransMatrixd	getMatrix(int index)	Obtain the transformation vector and convert to the matrix representation.
TransMatrixd	getMatrix(int index, float sample)
TransMatrixd	getMatrixHalf(int index)	Obtain the transformation vector, scale it by 0.5, then convert it to transformation matrix.
TransMatrixd	getMatrixHalf(int index, float sample)	Obtain the transformation vector, scale it by 0.5, then convert it to transformation matrix and scale the translations by sample parameter.
TransMatrixd	getMatrixMidsagittal(int index)	Accessor that returns the matrix representing the best tranformation.
TransMatrixd	getMatrixMidsagittal(int index, float sample)	Accessor that returns the matrix representing the best tranformation.
int	getMaxIterations()
double[]	getPoint(int index)	Return the full transformation vector.
double[]	getPoint(int index, float sample)	Obtain the transformation vector and adjust its translation by sample parameters.
Vectornd[]	getPoints()	Return an array of transformation vector.
final double	measureCost(double[] point)
void	runAlgorithm()	Actually runs the algorithm.
void	runELSUNC()	Runs ELSUNC along one dimension at a time as long as the costFunction improves during one cycle of runs along every dimension.
private void	runLM()	Runs Levneberg-Marqaurdt along one dimension at a time as long as the costFunction improves during one cycle of runs along every dimension.
private void	runNL2sol()
static final double[]	scalePoint(double[] point, double scale)	Scale the point by scale parameter and store it into another vector.
final double[]	scalePoint(int index, double scale)	Obtain the transformation vector and make a copy, then scale it by scale parameter.
void	setMaxIterations(int max)	Accessor that sets the maximum number of iterations.
void	setPoints(Vectornd[] points)	Sets the transformation vectors.
void	updatePoint(double[] point, double cost, Vectornd v)

Methods inherited from class gov.nih.mipav.model.algorithms.AlgorithmBase

actionPerformed, addListener, addProgressChangeListener, calculateImageSize, calculatePrincipleAxis, computeElapsedTime, computeElapsedTime, convertIntoFloat, delinkProgressToAlgorithm, delinkProgressToAlgorithmMulti, displayError, errorCleanUp, finalize, fireProgressStateChanged, fireProgressStateChanged, fireProgressStateChanged, fireProgressStateChanged, fireProgressStateChanged, generateProgressValues, getDestImage, getElapsedTime, getMask, getMaxProgressValue, getMinProgressValue, getNumberOfThreads, getProgress, getProgressChangeListener, getProgressChangeListeners, getProgressModulus, getProgressStep, getProgressValues, getSrcImage, isCompleted, isImage25D, isMultiThreadingEnabled, isRunningInSeparateThread, isThreadStopped, linkProgressToAlgorithm, linkProgressToAlgorithm, makeProgress, notifyListeners, removeListener, removeProgressChangeListener, run, setCompleted, setImage25D, setMask, setMaxProgressValue, setMinProgressValue, setMultiThreadingEnabled, setNumberOfThreads, setProgress, setProgressModulus, setProgressStep, setProgressValues, setProgressValues, setRunningInSeparateThread, setSrcImage, setStartTime, setThreadStopped, startMethod, windowActivated, windowClosed, windowClosing, windowDeactivated, windowDeiconified, windowIconified, windowOpened

Methods inherited from class java.lang.Thread

activeCount, checkAccess, clone, countStackFrames, currentThread, dumpStack, enumerate, getAllStackTraces, getContextClassLoader, getDefaultUncaughtExceptionHandler, getId, getName, getPriority, getStackTrace, getState, getThreadGroup, getUncaughtExceptionHandler, holdsLock, interrupt, interrupted, isAlive, isDaemon, isInterrupted, isVirtual, join, join, join, join, ofPlatform, ofVirtual, onSpinWait, resume, setContextClassLoader, setDaemon, setDefaultUncaughtExceptionHandler, setName, setPriority, setUncaughtExceptionHandler, sleep, sleep, sleep, start, startVirtualThread, stop, suspend, threadId, toString, yield

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Details

ELSUNC

private final int ELSUNC

See Also:

• Constant Field Values

LEVENBERG_MARQUARDT

private final int LEVENBERG_MARQUARDT

See Also:

• Constant Field Values

NL2SOL

private final int NL2SOL

See Also:

• Constant Field Values

searchAlgorithm

private int searchAlgorithm

nDims

private int nDims

Degress of freedom.

costFunction

private AlgorithmOptimizeFunctionBase costFunction

Cost function called to measure cost - 1D.

OARTolerance

private double[] OARTolerance

Array of tolerances for each dimension.

maxIterations

private int maxIterations

The maximum number of iterations the optimization allows.

parent

private AlgorithmBase parent

Parent algorithm that called this optimization.

toOrigin

private TransMatrixd toOrigin

The transformation matrix to the origin of the input image.

fromOrigin

private TransMatrixd fromOrigin

The transformation matrix from the origin of the input image.

points

private Vectornd[] points

Array used to hold the initial points, final points and costs

functionAtBest

private double functionAtBest

The cost of the function at the best minimum.

minFunctionAtBest

private double minFunctionAtBest

eModel

AlgorithmELSUNCOpt3D.FitOAR3DELSUNCModel eModel

nModel

AlgorithmELSUNCOpt3D.FitOAR3DNL2solModel nModel

lmModel

AlgorithmELSUNCOpt3D.FitOAR3DLMModel lmModel

status

private int status

start

private double[] start

Point that was initially passed into function.

epsilon

private double epsilon

huge

private double huge

Constructor Details

AlgorithmELSUNCOpt3D

public AlgorithmELSUNCOpt3D(AlgorithmBase parent,
 WildMagic.LibFoundation.Mathematics.Vector3f com,
 int degreeOfFreedom,
 AlgorithmOptimizeFunctionBase costFunc,
 double[] tols,
 int maxIter,
 int searchAlgorithm)

Constructs a new algorithm with the given centers of mass (needed for setting the transformations), the given cost function (which was constructed with the proper images), the initial point we're looking at, some tolerance within that point to look for the minimum, and the maximum number of iterations.

Parameters:

parent - Algorithm that called this optimization.

com - Center of Mass of the input image.

degreeOfFreedom - Degree of freedom for transformation (must be 3, 4, 6, 7, 9, or 12).

costFunc - Cost function to use.

tols - Tolerance for each dimension (tols.length == degreeOfFreedom).

maxIter - Maximum number of iterations.

searchAlgorithm - ELSUNC, LEVENBERG_MARQUARDT, or NL2SOL;

Method Details

disposeLocal

public void disposeLocal()

Sets everything to null and prepares this class for destruction.

getPoints

public Vectornd[] getPoints()

Return an array of transformation vector. The meanings of those transformation vector are as following: the initial transformation vector: before algorithm is performed the final transformation vector: after algorithm was performed.

Returns:

setPoints

public void setPoints(Vectornd[] points)

Sets the transformation vectors.

Parameters:

points - the transformation vectors.

runAlgorithm

public void runAlgorithm()

Description copied from class: AlgorithmBase

Actually runs the algorithm. Implemented by inheriting algorithms.

Specified by:

runAlgorithm in class AlgorithmBase

runELSUNC

public void runELSUNC()

Runs ELSUNC along one dimension at a time as long as the costFunction improves during one cycle of runs along every dimension.

runNL2sol

private void runNL2sol()

dfault

private void dfault(int[] iv,
 double[] v)

runLM

private void runLM()

Runs Levneberg-Marqaurdt along one dimension at a time as long as the costFunction improves during one cycle of runs along every dimension.

constructPoint

public double[] constructPoint(double[] defaultPoint,
 double[] point)

Construct a full 12-dimension transformation vector from the partial transformation vector. For missing values in point, the values in defaultPoint will be used. Different degree of freedom has different meanings: 3: only 3 translations 4: 3 translation and global scaling 6: 3 rotations and translations 7: 3 rotations and translations, and global scaling 9: 3 rotations, translations and scalings 12: 3 rotations, translations, scalings and skewings.

Parameters:

defaultPoint - a default full 12-dimension transformation vector.

point - a partial or full transformation vector.

Returns:

a full transformation vector.

getFinal

public double[] getFinal(double[] point)

Accessor that returns the final point with translations, rotations, scales, and skews representing the best tranformation.

Returns:

vector representing the best transformation in terms of translations, rotations, scales, and skews.

convertToMatrix

public TransMatrixd convertToMatrix(TransMatrixd toOrigin,
 TransMatrixd fromOrigin,
 double[] vector)

Convert a 12-dimension transformation vector to a 4x4 transformation matrix.

Parameters:

vector - a 12-dimension transformation vector including 3 rotations, translations, scalings and skews.

Returns:

a 4x4 transformation matrix

extractPoint

public double[] extractPoint(double[] startPoint)

Extract the partial or full transformation vector from the start transformation vector, which will be optimized.

Parameters:

startPoint - the start full 12-dimension transformation vector.

Returns:

the partial or full transformation vector which will be optimized.

getMatrix

public TransMatrixd getMatrix(int index,
 float sample)

See Also:

• AlgorithmPowellOptBase.getMatrix(int, float)

getMatrix

public final TransMatrixd getMatrix(int index)

Obtain the transformation vector and convert to the matrix representation.

Parameters:

index - the index of transformation vector.

Returns:

the transformation matrix

convertToMatrix

public TransMatrixd convertToMatrix(double[] vector)

Convert a transformation vector to a transformation matrix.

Parameters:

vector - a transformation vector.

Returns:

a transformation matrix

measureCost

public final double measureCost(double[] point)

getCost

public final double getCost(int index)

Returns the cost for the transformation vector.

Parameters:

index - the index of transformation vector.

Returns:

the cost for the transformation vector.

getPoint

public double[] getPoint(int index)

Return the full transformation vector.

Parameters:

index - the index of the transformation vector.

Returns:

the full transformation vector.

getPoint

public double[] getPoint(int index,
 float sample)

Obtain the transformation vector and adjust its translation by sample parameters.

Parameters:

index - the index of transformation vector.

sample - the translation scaling parameter.

Returns:

the translation scaled transformation vector.

getMatrix

public TransMatrixd getMatrix(double[] point,
 float sample)

See Also:

• AlgorithmPowellOptBase.getMatrix(int, float)

adjustTranslation

public void adjustTranslation(TransMatrixd mat,
 float sample)

See Also:

• invalid reference

  AlgorithmPowellOptBase#adjustTranslation(TransMatrix, float)

getMatrixHalf

public TransMatrixd getMatrixHalf(int index)

Obtain the transformation vector, scale it by 0.5, then convert it to transformation matrix.

Parameters:

index - the index of transformation vector.

Returns:

the scaled transformation matrix.

scalePoint

public static final double[] scalePoint(double[] point,
 double scale)

Scale the point by scale parameter and store it into another vector.

Parameters:

point - the transformation vector.

scale - the scale parameter.

Returns:

the scaled transformation vector

scalePoint

public final double[] scalePoint(int index,
 double scale)

Obtain the transformation vector and make a copy, then scale it by scale parameter.

Parameters:

index - the index of transformation vector

scale - the scale parameter

Returns:

the new scaled transformation vector.

getMatrixHalf

public TransMatrixd getMatrixHalf(int index,
 float sample)

Obtain the transformation vector, scale it by 0.5, then convert it to transformation matrix and scale the translations by sample parameter.

Parameters:

index - the index of transformation vector.

sample - the translation scaling parameter.

Returns:

the scaled transformation matrix.

convertToMidsagittal

public double[] convertToMidsagittal(double[] point)

Converts a full transformation vector into a midsagittal transformation vector.

Parameters:

point - the full 12-dimension transformation vector.

Returns:

the transformation vector complying with midsagittal alignment algorithm.

copyPoint

public static final double[] copyPoint(double[] point)

Make a copy of the transformation vector

Parameters:

point - a transformation vector.

Returns:

the copy of the transformation vector.

getMatrixMidsagittal

public TransMatrixd getMatrixMidsagittal(int index)

Accessor that returns the matrix representing the best tranformation. This transformation contains only the z rotation and the x and y translation, to be used in the midsagittal alignment algorithm.

Returns:

matrix representing the best transformation's z rot and x and y trans.

getMatrixMidsagittal

public TransMatrixd getMatrixMidsagittal(int index,
 float sample)

Accessor that returns the matrix representing the best tranformation. The passed in parameter represents the resolution (same in all directions and for both input and reference images, since resampled isotropically). Since the optimization was done in pixel space, not millimeter space, the translation parameters need to be scaled by the sample value. This transformation contains only the z rotation and the x and y translation, to be used in the midsagittal alignment algorithm.

Parameters:

sample - the voxel resolution

Returns:

matrix representing the best transformation's z rot and x and y trans.

updatePoint

public void updatePoint(double[] point,
 double cost,
 Vectornd v)

See Also:

• AlgorithmPowellOptBase.updatePoint(double[], double, Vectornd)

setMaxIterations

public void setMaxIterations(int max)

Accessor that sets the maximum number of iterations.

Parameters:

max - The max number of iterations.

getMaxIterations

public int getMaxIterations()