[Insight-users] Poor results with the BSplineDeformableTransform

[motes motes]

I have made a know deformation of lena using the BSplineWarping1.cxx
with this input (12*12 nodes on the image. Since I use cubic BSpline
this gives a total of 15*15 nodes):


    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0    -10  20     10  20       0   0      0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0    -10  -0    10  -20       0   0      0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0

    0   0       0   0       0   0        0   0       0   0        0
0       0   0      0   0      0   0       0   0       0   0        0
0       0   0        0   0       0   0


In the image registration method I use the following components:


typedef itk::BSplineDeformableTransform<CoordinateRepType, Dimension,
SplineOrder >          TransformType;
typedef itk::MeanSquaresImageToImageMetric< FixedImageType,
MovingImageType >              MetricType;
typedef itk::LinearInterpolateImageFunction< MovingImageType, double >
                                InterpolatorType;
typedef itk::RegularStepGradientDescentOptimizer	
                                    OptimizerType;

I setup the  transform as shown below (8*8 nodes on the image. Since I
use cubic BSpline this gives a total of 12*12 nodes):

    gridSizeOnImage.Fill(8);
    gridBorderSize.Fill(3);    // Border for spline order = 3 ( 1
lower, 2 upper )

    //gridSizeOnImage.Fill( 5 );
    //gridBorderSize.Fill( 3 );    // Border for spline order = 3 ( 1
lower, 2 upper )

    totalGridSize = gridSizeOnImage + gridBorderSize;
    bsplineRegion.SetSize( totalGridSize );
    for(unsigned int r=0; r<Dimension; r++) {
      spacing[r] *= static_cast<double>(fixedImageSize[r] - 1)  /
        static_cast<double>(gridSizeOnImage[r] - 1);
    }
    SpacingType gridOriginOffset = gridDirection * spacing;
    OriginType gridOrigin = origin - gridOriginOffset;
    transform->SetGridSpacing(spacing);
    transform->SetGridOrigin(gridOrigin);
    transform->SetGridRegion(bsplineRegion);
    transform->SetGridDirection(gridDirection);
    const unsigned int numberOfParameters = transform->GetNumberOfParameters();
    ParametersType parameters(numberOfParameters);
    parameters.Fill(0.0);
    transform->SetParameters(parameters);
    registration->SetInitialTransformParameters(transform->GetParameters());

After 500 iterations the metric outputs 35.85 and the result can be seen below:

http://37133.vs.webtropia.com/apache2-default/test3.jpg


But is that not a rather poor result when taking into consideration
that it has run for 500 iterations? Or is it because I used a finer
resolution when creating the moving image than when running the image
registration method?