A new class shall be created which can generate a (3D) image, using multiple stacked Gauss functions to define pixel intensities. The purpose of this class is to create test images for testing of the statistics module. By using 3D Gauss functions, an image resembling typical PET scans can be created.
The class should also allow to calculate the "hot spot" of the generated image, i.e. a spheric region of a defined size in the image which has the maximum mean value.
| Status | Assigned | Task | ||
|---|---|---|---|---|
| Resolved | None | T16236 Enhance mitkImageStatisticsCalculator with hotspot search | ||
| Resolved | None | T16396 Implement ITK class for generating 3D images defined by multiple Gauss functions |
I just rebased the code onto a different branching point in order to make it merge easier with the changes of T16236.
The class name is MultiGaussianImageSource and can be found in the directory MINT\Mint_Bin\CMakeExternals\Source\MITK\Modules\ImageStatistics.
This class generates a image, which pixel intensities are the values of a multigauss function (a sum of 3D gauss functions). The class allows calculating a spherical region of defined size which has a maximal mean value and returns the position and the mean value. Furthermore it can calculate the maximal und minimal pixel intensities and whose indices in the founded sphere.
The test programm (mitkMultiGaussianTest.cpp) creates a .xml file with the following information:
This test program has an 14 inputs:
The input list could be extended, for example with a variable region of interest or a min/max expectancy value (here is the min/max expectancy value equal the size of the image +/- 7).
The class name is MultiGaussianImageSource and can be found in the directory MINT\Mint_Bin\CMakeExternals\Source\MITK\Modules\ImageStatistics.
This class generates a image, which pixel intensities are the values of a multigauss function (a sum of 3D gauss functions). The class allows calculating a spherical region of defined size which has a maximal mean value and returns the position and the mean value. Furthermore it can calculate the maximal und minimal pixel intensities and whose indices in the founded sphere.
The test programm (mitkMultiGaussianTest.cpp) creates a .xml file with the following information:
This test program has an 14 inputs:
The input list could be extended, for example with a variable region of interest or a min/max expectancy value (here is the min/max expectancy value equal the size of the image +/- 7).
Added a subdivision of the sphere in cuboids wiht octrees to calculate analytical the integral over gaussian with the value tabel of the normal distribution (http://www.stat.tamu.edu/~lzhou/stat302/standardnormaltable.pdf).
We needed a cuboid because the antiderivativ can not be given in an explicit way and therefore we are not able to subsitude in the integral with spherical coordinates.
Added a pinput parameter entireHotSpotInImage, which determines whether the whole sphere is in the image included ( = 1 ) or only the midpoint of the sphere is inside the image.
For generating the gaussian image change the way we calculate the value in each voxel:
instead of taking the value at the voxel's midpoint we calculate the mean value in the voxel with an integral of the gaussians over the voxel's boundary.