void mitk::CreateDistanceImageFromSurfaceFilter::CreateDistanceImage()
{
typedef itk::Image<double, 3> DistanceImageType;
typedef itk::ImageRegionIteratorWithIndex<DistanceImageType> ImageIterator;
typedef itk::NeighborhoodIterator<DistanceImageType> NeighborhoodImageIterator;
DistanceImageType::Pointer distanceImg = DistanceImageType::New();
//Determin the bounding box of the delineated contours
double xmin = m_Centers.at(0)[0];
double ymin = m_Centers.at(0)[1];
double zmin = m_Centers.at(0)[2];
double xmax = m_Centers.at(0)[0];
double ymax = m_Centers.at(0)[1];
double zmax = m_Centers.at(0)[2];
for (unsigned int i = 1; i < m_Centers.size(); i++)
{
if (xmin > m_Centers.at(i)[0])
{
xmin = m_Centers.at(i)[0];
}
if (ymin > m_Centers.at(i)[1])
{
ymin = m_Centers.at(i)[1];
}
if (zmin > m_Centers.at(i)[2])
{
zmin = m_Centers.at(i)[2];
}
if (xmax < m_Centers.at(i)[0])
{
xmax = m_Centers.at(i)[0];
}
if (ymax < m_Centers.at(i)[1])
{
ymax = m_Centers.at(i)[1];
}
if (zmax < m_Centers.at(i)[2])
{
zmax = m_Centers.at(i)[2];
}
}
Vector3D extentMM;
extentMM[0] = xmax - xmin + 2;
extentMM[1] = ymax - ymin + 2;
extentMM[2] = zmax - zmin + 2;
//Shifting the distance image's offest to achieve an exact distance calculation
xmin = xmin - 2;
ymin = ymin - 2;
zmin = zmin - 2;
/*
Now create an empty distance image. The create image will always have the same size, independent from
the original image (e.g. always consists of 500000 pixels) and will have an isotropic spacing.
The spacing is calculated like the following:
The image's volume = 500000 Pixels = extentX*spacing*extentY*spacing*extentZ*spacing
So the spacing is: spacing = ( 500000 / extentX*extentY*extentZ )^(1/3)
*/
double basis = (extentMM[0]*extentMM[1]*extentMM[2]) / m_DistanceImageVolume;
double exponent = 1.0/3.0;
double distImgSpacing = pow(basis, exponent);
int tempSpacing = (distImgSpacing+0.05)*10;
m_DistanceImageSpacing = (double)tempSpacing/10.0;
unsigned int numberOfXPixel = extentMM[0] / m_DistanceImageSpacing;
unsigned int numberOfYPixel = extentMM[1] / m_DistanceImageSpacing;
unsigned int numberOfZPixel = extentMM[2] / m_DistanceImageSpacing;
DistanceImageType::SizeType size;
//Increase the distance image's size a little bit to achieve an exact distance calculation
size[0] = numberOfXPixel + 5;
size[1] = numberOfYPixel + 5;
size[2] = numberOfZPixel + 5;
DistanceImageType::IndexType start;
start[0] = 0;
start[1] = 0;
start[2] = 0;
DistanceImageType::RegionType lpRegion;
lpRegion.SetSize(size);
lpRegion.SetIndex(start);
distanceImg->SetRegions( lpRegion );
distanceImg->SetSpacing( m_DistanceImageSpacing );
distanceImg->Allocate();
//First of all the image is initialized with the value 10 for each pixel
distanceImg->FillBuffer(10);
/*
Now we must caculate the distance for each pixel. But instead of calculating the distance value
for all of the image's pixels we proceed similar to the region growing algorithm:
1. Take the first pixel from the narrowband_point_list and calculate the distance for each neighbor (6er)
2. If the current index's distance value is below a certain threshold push it into the list
3. Next iteration take the next index from the list and start with 1. again
This is done until the narrowband_point_list is empty.
*/
std::queue<DistanceImageType::IndexType> narrowbandPoints;
PointType currentPoint = m_Centers.at(0);
double distance = this->CalculateDistanceValue(currentPoint);
DistanceImageType::IndexType currentIndex;
currentIndex[0] = ( currentPoint[0]-xmin ) / m_DistanceImageSpacing;
currentIndex[1] = ( currentPoint[1]-ymin ) / m_DistanceImageSpacing;
currentIndex[2] = ( currentPoint[2]-zmin ) / m_DistanceImageSpacing;
narrowbandPoints.push(currentIndex);
distanceImg->SetPixel(currentIndex, distance);
NeighborhoodImageIterator::RadiusType radius;
radius.Fill(1);
NeighborhoodImageIterator nIt(radius, distanceImg, distanceImg->GetLargestPossibleRegion());
unsigned int relativeNbIdx[] = {4, 10, 12, 14, 16, 22};
bool isInBounds = false;
while ( !narrowbandPoints.empty() )
{
nIt.SetLocation(narrowbandPoints.front());
narrowbandPoints.pop();
for (int i = 0; i < 6; i++)
{
nIt.GetPixel(relativeNbIdx[i], isInBounds);
if( isInBounds && nIt.GetPixel(relativeNbIdx[i]) == 10)
{
currentIndex = nIt.GetIndex(relativeNbIdx[i]);
currentPoint[0] = currentIndex[0]*m_DistanceImageSpacing + xmin;
currentPoint[1] = currentIndex[1]*m_DistanceImageSpacing + ymin;
currentPoint[2] = currentIndex[2]*m_DistanceImageSpacing + zmin;
distance = this->CalculateDistanceValue(currentPoint);
if ( abs(distance) <= m_DistanceImageSpacing*2 )
{
nIt.SetPixel(relativeNbIdx[i], distance);
narrowbandPoints.push(currentIndex);
}
}
}
}
ImageIterator imgRegionIterator (distanceImg, distanceImg->GetLargestPossibleRegion());
imgRegionIterator.GoToBegin();
double prevPixelVal = 1;
unsigned int _size[3] = { (unsigned int)(size[0] - 1), (unsigned int)(size[1] - 1), (unsigned int)(size[2] - 1) };
//Set every pixel inside the surface to -10 except the edge point (so that the received surface is closed)
while (!imgRegionIterator.IsAtEnd()) {
if ( imgRegionIterator.Get() == 10 && prevPixelVal < 0 )
{
while (imgRegionIterator.Get() == 10)
{
if (imgRegionIterator.GetIndex()[0] == _size[0] || imgRegionIterator.GetIndex()[1] == _size[1] || imgRegionIterator.GetIndex()[2] == _size[2]
|| imgRegionIterator.GetIndex()[0] == 0U || imgRegionIterator.GetIndex()[1] == 0U || imgRegionIterator.GetIndex()[2] == 0U )
{
imgRegionIterator.Set(10);
prevPixelVal = 10;
++imgRegionIterator;
break;
}
else
{
imgRegionIterator.Set(-10);
++imgRegionIterator;
prevPixelVal = -10;
}
}
}
else if (imgRegionIterator.GetIndex()[0] == _size[0] || imgRegionIterator.GetIndex()[1] == _size[1] || imgRegionIterator.GetIndex()[2] == _size[2]
|| imgRegionIterator.GetIndex()[0] == 0U || imgRegionIterator.GetIndex()[1] == 0U || imgRegionIterator.GetIndex()[2] == 0U)
{
imgRegionIterator.Set(10);
prevPixelVal = 10;
++imgRegionIterator;
}
else {
prevPixelVal = imgRegionIterator.Get();
++imgRegionIterator;
}
}
Image::Pointer resultImage = this->GetOutput();
Point3D origin;
origin[0] = xmin;
origin[1] = ymin;
origin[2] = zmin;
CastToMitkImage(distanceImg, resultImage);
resultImage->GetGeometry()->SetOrigin(origin);
resultImage->SetOrigin(origin);
}
void mitk::CreateDistanceImageFromSurfaceFilter::CreateDistanceImage()
{
DistanceImageType::Pointer distanceImg = DistanceImageType::New();
// Determine the bounds of the input points in index- and world-coordinates
DistanceImageType::PointType minPointInWorldCoordinates, maxPointInWorldCoordinates;
DistanceImageType::IndexType minPointInIndexCoordinates, maxPointInIndexCoordinates;
DetermineBounds( minPointInWorldCoordinates, maxPointInWorldCoordinates,
minPointInIndexCoordinates, maxPointInIndexCoordinates );
// Calculate the extent of the region that contains all given points in MM.
// To do this, we take the difference between the maximal and minimal
// index-coordinates (must not be less than 1) and multiply it with the
// spacing of the reference-image.
Vector3D extentMM;
for (unsigned int dim = 0; dim < 3; ++dim)
{
extentMM[dim] = (int)
(
(std::max( std::abs(maxPointInIndexCoordinates[dim] - minPointInIndexCoordinates[dim]),
(DistanceImageType::IndexType::IndexValueType) 1
) + 1.0) // (max-index - min-index)+1 because the pixels between index 3 and 5 cover 2+1=3 pixels (pixel 3,4, and 5)
* m_ReferenceImage->GetSpacing()[dim]
) + 1; // (int) ((...) + 1) -> we round up to the next BIGGER int value
}
/*
* Now create an empty distance image. The create image will always have the same sizeOfRegion, independent from
* the original image (e.g. always consists of 500000 pixels) and will have an isotropic spacing.
* The spacing is calculated like the following:
* The image's volume = 500000 Pixels = extentX*spacing*extentY*spacing*extentZ*spacing
* So the spacing is: spacing = ( 500000 / extentX*extentY*extentZ )^(1/3)
*/
double basis = (extentMM[0]*extentMM[1]*extentMM[2]) / m_DistanceImageVolume;
double exponent = 1.0/3.0;
double distImgSpacing = pow(basis, exponent);
int tempSpacing = (distImgSpacing+0.05)*10;
m_DistanceImageSpacing = (double)tempSpacing/10.0;
// calculate the number of pixels of the distance image for each direction
unsigned int numberOfXPixel = extentMM[0] / m_DistanceImageSpacing;
unsigned int numberOfYPixel = extentMM[1] / m_DistanceImageSpacing;
unsigned int numberOfZPixel = extentMM[2] / m_DistanceImageSpacing;
// We increase the sizeOfRegion by 4 as we decrease the origin by 2 later.
// This expansion of the region is necessary to achieve a complete
// interpolation.
DistanceImageType::SizeType sizeOfRegion;
sizeOfRegion[0] = numberOfXPixel + 4;
sizeOfRegion[1] = numberOfYPixel + 4;
sizeOfRegion[2] = numberOfZPixel + 4;
// The region starts at index 0,0,0
DistanceImageType::IndexType initialOriginAsIndex;
initialOriginAsIndex.Fill(0);
DistanceImageType::PointType originAsWorld = minPointInWorldCoordinates;
DistanceImageType::RegionType lpRegion;
lpRegion.SetSize(sizeOfRegion);
lpRegion.SetIndex(initialOriginAsIndex);
// We initialize the itk::Image with
// * origin and direction to have it correctly placed and rotated in the world
// * the largest possible region to set the extent to be calculated
// * the isotropic spacing that we have calculated above
distanceImg->SetOrigin( originAsWorld );
distanceImg->SetDirection( m_ReferenceImage->GetDirection() );
distanceImg->SetRegions( lpRegion );
distanceImg->SetSpacing( m_DistanceImageSpacing );
distanceImg->Allocate();
//First of all the image is initialized with the value 10 for each pixel
distanceImg->FillBuffer(10);
// Now we move the origin of the distanceImage 2 index-Coordinates
// in all directions
DistanceImageType::IndexType originAsIndex;
distanceImg->TransformPhysicalPointToIndex( originAsWorld, originAsIndex );
originAsIndex[0] -= 2;
originAsIndex[1] -= 2;
originAsIndex[2] -= 2;
distanceImg->TransformIndexToPhysicalPoint( originAsIndex, originAsWorld );
distanceImg->SetOrigin( originAsWorld );
/*
* Now we must calculate the distance for each pixel. But instead of calculating the distance value
* for all of the image's pixels we proceed similar to the region growing algorithm:
*
* 1. Take the first pixel from the narrowband_point_list and calculate the distance for each neighbor (6er)
* 2. If the current index's distance value is below a certain threshold push it into the list
* 3. Next iteration take the next index from the list and originAsIndex with 1. again
*
* This is done until the narrowband_point_list is empty.
*/
std::queue<DistanceImageType::IndexType> narrowbandPoints;
PointType currentPoint = m_Centers.at(0);
double distance = this->CalculateDistanceValue(currentPoint);
// create itk::Point from vnl_vector
DistanceImageType::PointType currentPointAsPoint;
currentPointAsPoint[0] = currentPoint[0];
currentPointAsPoint[1] = currentPoint[1];
currentPointAsPoint[2] = currentPoint[2];
// Transform the input point in world-coordinates to index-coordinates
DistanceImageType::IndexType currentIndex;
distanceImg->TransformPhysicalPointToIndex( currentPointAsPoint, currentIndex );
assert( lpRegion.IsInside(currentIndex) ); // we are quite certain this should hold
narrowbandPoints.push(currentIndex);
distanceImg->SetPixel(currentIndex, distance);
NeighborhoodImageIterator::RadiusType radius;
radius.Fill(1);
NeighborhoodImageIterator nIt(radius, distanceImg, distanceImg->GetLargestPossibleRegion());
unsigned int relativeNbIdx[] = {4, 10, 12, 14, 16, 22};
bool isInBounds = false;
while ( !narrowbandPoints.empty() )
{
nIt.SetLocation(narrowbandPoints.front());
narrowbandPoints.pop();
unsigned int* relativeNb = &relativeNbIdx[0];
for (int i = 0; i < 6; i++)
{
nIt.GetPixel(*relativeNb, isInBounds);
if( isInBounds && nIt.GetPixel(*relativeNb) == 10)
{
currentIndex = nIt.GetIndex(*relativeNb);
// Transform the currently checked point from index-coordinates to
// world-coordinates
distanceImg->TransformIndexToPhysicalPoint( currentIndex, currentPointAsPoint );
// create a vnl_vector
currentPoint[0] = currentPointAsPoint[0];
currentPoint[1] = currentPointAsPoint[1];
currentPoint[2] = currentPointAsPoint[2];
// and check the distance
distance = this->CalculateDistanceValue(currentPoint);
if ( abs(distance) <= m_DistanceImageSpacing )
{
nIt.SetPixel(*relativeNb, distance);
narrowbandPoints.push(currentIndex);
}
}
relativeNb++;
}
}
// Fist we set the border slices of the image to value 1000 so that we can perform a
// region growing afterwards starting from the middle of the image
DistanceImageType::SizeType reqSize;
reqSize[0] = distanceImg->GetLargestPossibleRegion().GetSize()[0];
reqSize[1] = distanceImg->GetLargestPossibleRegion().GetSize()[1];
reqSize[2] = 1;
DistanceImageType::IndexType reqStart;
reqStart[0] = 0;
reqStart[1] = 0;
reqStart[2] = 0;
DistanceImageType::RegionType reqRegion;
reqRegion.SetSize(reqSize);
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
reqStart[0] = 0;
reqStart[1] = 0;
reqStart[2] = distanceImg->GetLargestPossibleRegion().GetSize()[2]-1;
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
reqSize[0] = 1;
reqSize[1] = distanceImg->GetLargestPossibleRegion().GetSize()[1];
reqSize[2] = distanceImg->GetLargestPossibleRegion().GetSize()[2];;
reqStart[0] = 0;
reqStart[1] = 0;
reqStart[2] = 0;
reqRegion.SetSize(reqSize);
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
reqStart[0] = distanceImg->GetLargestPossibleRegion().GetSize()[0]-1;
reqStart[1] = 0;
reqStart[2] = 0;
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
reqSize[0] = distanceImg->GetLargestPossibleRegion().GetSize()[0];
reqSize[1] = 1;
reqSize[2] = distanceImg->GetLargestPossibleRegion().GetSize()[2];;
reqStart[0] = 0;
reqStart[1] = 0;
reqStart[2] = 0;
reqRegion.SetSize(reqSize);
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
reqStart[0] = 0;
reqStart[1] = distanceImg->GetLargestPossibleRegion().GetSize()[1]-1;
reqStart[2] = 0;
reqRegion.SetIndex(reqStart);
this->FillImageRegion(reqRegion, 1000, distanceImg);
// Now we make some kind of region growing from the middle of the image to set all
// inner pixels to -10. In this way we assure to extract a valid surface
NeighborhoodImageIterator nIt2(radius, distanceImg, distanceImg->GetLargestPossibleRegion());
currentIndex[0] = distanceImg->GetLargestPossibleRegion().GetSize()[0]*0.5;
currentIndex[1] = distanceImg->GetLargestPossibleRegion().GetSize()[1]*0.5;
currentIndex[2] = distanceImg->GetLargestPossibleRegion().GetSize()[2]*0.5;
narrowbandPoints.push(currentIndex);
distanceImg->SetPixel(currentIndex, -10);
while ( !narrowbandPoints.empty() )
{
nIt2.SetLocation(narrowbandPoints.front());
narrowbandPoints.pop();
for (int i = 0; i < 6; i++)
{
if( nIt2.GetPixel(relativeNbIdx[i]) == 10)
{
currentIndex = nIt2.GetIndex(relativeNbIdx[i]);
nIt2.SetPixel(relativeNbIdx[i], -10);
narrowbandPoints.push(currentIndex);
}
}
}
Image::Pointer resultImage = this->GetOutput();
// Cast the created distance-Image from itk::Image to the mitk::Image
// that is our output.
CastToMitkImage(distanceImg, resultImage);
}