//#################### PRIVATE METHODS ####################
void SpineIdentifier3D::execute_impl()
{
set_status("Identifying the spine...");
VolumeIPFMultiFeatureSelection_Ptr multiFeatureSelection = get_multi_feature_selection();
// Step 1: Filter for spine.
std::list<PFNodeID> nodes = filter_branch_nodes(boost::bind(&SpineIdentifier3D::is_spine, this, _1, _2));
// Step 2: Convert the partition forest candidate nodes to positions in the volume.
std::set<int> leaves;
for(std::list<PFNodeID>::const_iterator it=nodes.begin(), iend=nodes.end(); it!=iend; ++it)
{
std::deque<int> leafIndices = volume_ipf()->receptive_region_of(*it);
for(std::deque<int>::const_iterator jt=leafIndices.begin(), jend=leafIndices.end(); jt!=jend; ++jt)
{
leaves.insert(*jt);
}
}
std::list<itk::Index<3> > positions;
for(std::set<int>::const_iterator it=leaves.begin(), iend=leaves.end(); it!=iend; ++it)
{
if(210 < volume_ipf()->leaf_properties(*it).grey_value())
{
positions.push_back(volume_ipf()->position_of_leaf(*it));
}
}
increment_progress();
// Step 3: Use the Fast Marching Method.
FastMarching<3> fm(volume_ipf()->leaf_layer()->gradient_magnitude_image(), positions);
positions = fm.get_shape_at_time(100.0);
increment_progress();
// Step 4: Convert the positions in the volume to partition forest selection.
leaves.clear();
for(std::list<itk::Index<3> >::const_iterator it=positions.begin(), iend=positions.end(); it!=iend; ++it)
{
leaves.insert(volume_ipf()->leaf_of_position(*it));
}
PartitionForestSelection_Ptr region(new PartitionForestSelectionT(volume_ipf(), leaves));
// Step 5: Mark the results as spine.
multiFeatureSelection->identify_selection(region, AbdominalFeature::VERTEBRA);
}
//#################### PUBLIC METHODS ####################
void CTIPFBuilder::execute()
{
typedef itk::Image<short,3> GradientMagnitudeImage;
typedef itk::Image<int,3> HounsfieldImage;
typedef itk::Image<float,3> RealImage;
typedef itk::Image<unsigned char,3> WindowedImage;
HounsfieldImage::Pointer hounsfieldImage = (*m_volume)->base_image();
//~~~~~~~
// STEP 1
//~~~~~~~
set_status("Preprocessing image...");
// Construct the windowed image.
WindowedImage::Pointer windowedImage = (*m_volume)->windowed_image(m_segmentationOptions.windowSettings);
if(is_aborted()) return;
// Cast the input image (whether Hounsfield or windowed) to make its pixels real-valued.
RealImage::Pointer realImage;
switch(m_segmentationOptions.inputType)
{
case CTSegmentationOptions::INPUTTYPE_HOUNSFIELD:
{
typedef itk::CastImageFilter<HounsfieldImage,RealImage> CastFilter;
CastFilter::Pointer castFilter = CastFilter::New();
castFilter->SetInput(hounsfieldImage);
castFilter->Update();
realImage = castFilter->GetOutput();
break;
}
case CTSegmentationOptions::INPUTTYPE_WINDOWED:
{
typedef itk::CastImageFilter<WindowedImage,RealImage> CastFilter;
CastFilter::Pointer castFilter = CastFilter::New();
castFilter->SetInput(windowedImage);
castFilter->Update();
realImage = castFilter->GetOutput();
break;
}
default:
{
throw Exception("Unknown CT segmentation input type"); // this should never happen
}
}
if(is_aborted()) return;
// Smooth this real image using anisotropic diffusion filtering.
typedef itk::GradientAnisotropicDiffusionImageFilter<RealImage,RealImage> ADFilter;
for(int i=0; i<m_segmentationOptions.adfIterations; ++i)
{
ADFilter::Pointer adFilter = ADFilter::New();
adFilter->SetInput(realImage);
adFilter->SetConductanceParameter(1.0);
adFilter->SetNumberOfIterations(1);
adFilter->SetTimeStep(0.0625);
adFilter->Update();
realImage = adFilter->GetOutput();
if(is_aborted()) return;
increment_progress();
}
// Calculate the gradient magnitude of the smoothed image.
typedef itk::GradientMagnitudeImageFilter<RealImage,GradientMagnitudeImage> GMFilter;
GMFilter::Pointer gmFilter = GMFilter::New();
gmFilter->SetInput(realImage);
gmFilter->SetUseImageSpacingOff();
gmFilter->Update();
GradientMagnitudeImage::Pointer gradientMagnitudeImage = gmFilter->GetOutput();
if(is_aborted()) return;
increment_progress();
//~~~~~~~
// STEP 2
//~~~~~~~
set_status("Running watershed...");
// Run the watershed algorithm on the gradient magnitude image.
typedef MeijsterRoerdinkWatershed<GradientMagnitudeImage::PixelType,3> WS;
WS ws(gradientMagnitudeImage, ITKImageUtil::make_6_connected_offsets());
if(is_aborted()) return;
increment_progress();
//~~~~~~~
// STEP 3
//~~~~~~~
set_status("Creating initial partition forest...");
boost::shared_ptr<CTImageLeafLayer> leafLayer(new CTImageLeafLayer(hounsfieldImage, windowedImage, gradientMagnitudeImage));
if(is_aborted()) return;
boost::shared_ptr<CTImageBranchLayer> lowestBranchLayer = IPF::make_lowest_branch_layer(leafLayer, ws.calculate_groups());
if(is_aborted()) return;
m_ipf.reset(new IPF(leafLayer, lowestBranchLayer));
increment_progress();
//~~~~~~~
// STEP 4
//~~~~~~~
set_status("Creating rooted MST for lowest branch layer...");
RootedMST<int> mst(*lowestBranchLayer);
if(is_aborted()) return;
increment_progress();
//~~~~~~~
// STEP 5
//~~~~~~~
set_status("Running waterfall...");
// Iteratively run a Nicholls waterfall pass on the MST until the forest is built.
typedef WaterfallPass<int>::Listener WaterfallPassListener;
NichollsWaterfallPass<int> waterfallPass;
boost::shared_ptr<WaterfallPassListener> listener = make_forest_building_waterfall_pass_listener(m_ipf);
waterfallPass.add_listener(listener);
while(mst.node_count() != 1 && m_ipf->highest_layer() < m_segmentationOptions.waterfallLayerLimit)
{
m_ipf->clone_layer(m_ipf->highest_layer());
if(is_aborted()) return;
waterfallPass.run(mst);
if(is_aborted()) return;
}
set_finished();
}
//#################### PUBLIC METHODS ####################
void CTLowestLayersBuilder::execute()
{
typedef itk::Image<short,3> GradientMagnitudeImage;
typedef itk::Image<int,3> HounsfieldImage;
typedef itk::Image<float,3> RealImage;
typedef itk::Image<unsigned char,3> WindowedImage;
HounsfieldImage::Pointer hounsfieldImage = (*m_volume)->base_image();
//~~~~~~~
// STEP 1
//~~~~~~~
set_status("Preprocessing image...");
// Construct the windowed image.
WindowedImage::Pointer windowedImage = (*m_volume)->windowed_image(m_segmentationOptions.windowSettings);
if(is_aborted()) return;
// Cast the input image (whether Hounsfield or windowed) to make its pixels real-valued.
RealImage::Pointer realImage;
switch(m_segmentationOptions.inputType)
{
case CTSegmentationOptions::INPUTTYPE_HOUNSFIELD:
{
typedef itk::CastImageFilter<HounsfieldImage,RealImage> CastFilter;
CastFilter::Pointer castFilter = CastFilter::New();
castFilter->SetInput(hounsfieldImage);
castFilter->Update();
realImage = castFilter->GetOutput();
break;
}
case CTSegmentationOptions::INPUTTYPE_WINDOWED:
{
typedef itk::CastImageFilter<WindowedImage,RealImage> CastFilter;
CastFilter::Pointer castFilter = CastFilter::New();
castFilter->SetInput(windowedImage);
castFilter->Update();
realImage = castFilter->GetOutput();
break;
}
default:
{
throw Exception("Unknown CT segmentation input type"); // this should never happen
}
}
if(is_aborted()) return;
// Smooth this real image using anisotropic diffusion filtering.
typedef itk::GradientAnisotropicDiffusionImageFilter<RealImage,RealImage> ADFilter;
for(int i=0; i<m_segmentationOptions.adfIterations; ++i)
{
ADFilter::Pointer adFilter = ADFilter::New();
adFilter->SetInput(realImage);
adFilter->SetConductanceParameter(1.0);
adFilter->SetNumberOfIterations(1);
adFilter->SetTimeStep(0.0625);
adFilter->Update();
realImage = adFilter->GetOutput();
if(is_aborted()) return;
increment_progress();
}
// Calculate the gradient magnitude of the smoothed image.
typedef itk::GradientMagnitudeImageFilter<RealImage,GradientMagnitudeImage> GMFilter;
GMFilter::Pointer gmFilter = GMFilter::New();
gmFilter->SetInput(realImage);
gmFilter->SetUseImageSpacingOff();
gmFilter->Update();
GradientMagnitudeImage::Pointer gradientMagnitudeImage = gmFilter->GetOutput();
if(is_aborted()) return;
increment_progress();
//~~~~~~~
// STEP 2
//~~~~~~~
set_status("Running watershed...");
// Run the watershed algorithm on the gradient magnitude image.
typedef MeijsterRoerdinkWatershed<GradientMagnitudeImage::PixelType,3> WS;
WS ws(gradientMagnitudeImage, ITKImageUtil::make_6_connected_offsets());
if(is_aborted()) return;
increment_progress();
//~~~~~~~
// STEP 3
//~~~~~~~
set_status("Creating lowest forest layers...");
m_leafLayer.reset(new CTImageLeafLayer(hounsfieldImage, windowedImage, gradientMagnitudeImage));
if(is_aborted()) return;
m_lowestBranchLayer = IPF::make_lowest_branch_layer(m_leafLayer, ws.calculate_groups());
if(is_aborted()) return;
set_finished();
}
