diff --git a/Modules/DiffusionImaging/FiberTracking/cmdapps/TractographyEvaluation/FitFibersToImage.cpp b/Modules/DiffusionImaging/FiberTracking/cmdapps/TractographyEvaluation/FitFibersToImage.cpp
index 3c7c93bfb9..68234fad4b 100755
--- a/Modules/DiffusionImaging/FiberTracking/cmdapps/TractographyEvaluation/FitFibersToImage.cpp
+++ b/Modules/DiffusionImaging/FiberTracking/cmdapps/TractographyEvaluation/FitFibersToImage.cpp
@@ -1,747 +1,738 @@
/*===================================================================
The Medical Imaging Interaction Toolkit (MITK)
Copyright (c) German Cancer Research Center,
Division of Medical and Biological Informatics.
All rights reserved.
This software is distributed WITHOUT ANY WARRANTY; without
even the implied warranty of MERCHANTABILITY or FITNESS FOR
A PARTICULAR PURPOSE.
See LICENSE.txt or http://www.mitk.org for details.
===================================================================*/
#include <mitkBaseData.h>
#include <mitkImageCast.h>
#include <mitkImageToItk.h>
#include <metaCommand.h>
#include <mitkCommandLineParser.h>
#include <usAny.h>
#include <mitkIOUtil.h>
#include <boost/lexical_cast.hpp>
#include <itksys/SystemTools.hxx>
#include <itkDirectory.h>
#include <mitkFiberBundle.h>
#include <mitkPreferenceListReaderOptionsFunctor.h>
#include <mitkDiffusionPropertyHelper.h>
#include <vnl/vnl_linear_system.h>
#include <Eigen/Dense>
#include <mitkStickModel.h>
#include <mitkBallModel.h>
#include <vigra/regression.hxx>
#include <itkImageFileWriter.h>
#include <itkImageDuplicator.h>
#include <itkMersenneTwisterRandomVariateGenerator.h>
#include <mitkPeakImage.h>
#include <vnl/algo/vnl_lbfgsb.h>
#include <vnl/vnl_sparse_matrix.h>
#include <vnl/vnl_sparse_matrix_linear_system.h>
#include <vnl/algo/vnl_lsqr.h>
#include <itkImageDuplicator.h>
#include <itkTimeProbe.h>
#include <random>
#include <itkParticleSwarmOptimizer.h>
#include <itkOnePlusOneEvolutionaryOptimizer.h>
#include <itkGradientDescentOptimizer.h>
#include <itkSPSAOptimizer.h>
using namespace std;
typedef itksys::SystemTools ist;
typedef itk::Point<float, 4> PointType4;
typedef itk::Image< float, 4 > PeakImgType;
-const double UPSCALE = 1.0;
vnl_vector_fixed<float,3> GetClosestPeak(itk::Index<4> idx, PeakImgType::Pointer peak_image , vnl_vector_fixed<float,3> fiber_dir, int& id, double& w )
{
int m_NumDirs = peak_image->GetLargestPossibleRegion().GetSize()[3]/3;
vnl_vector_fixed<float,3> out_dir; out_dir.fill(0);
float angle = 0.8;
for (int i=0; i<m_NumDirs; i++)
{
vnl_vector_fixed<float,3> dir;
idx[3] = i*3;
dir[0] = peak_image->GetPixel(idx);
idx[3] += 1;
dir[1] = peak_image->GetPixel(idx);
idx[3] += 1;
dir[2] = peak_image->GetPixel(idx);
float mag = dir.magnitude();
if (mag<mitk::eps)
continue;
dir.normalize();
float a = dot_product(dir, fiber_dir);
if (fabs(a)>angle)
{
angle = fabs(a);
w = angle;
if (a<0)
out_dir = -dir;
else
out_dir = dir;
out_dir *= mag;
id = i;
}
}
return out_dir;
}
class VnlCostFunction : public vnl_cost_function
{
public:
vnl_sparse_matrix_linear_system< double >* S;
vnl_sparse_matrix< double > m_A;
+ vnl_sparse_matrix< double > m_A_Ones; // matrix indicating active weights with 1
vnl_vector< double > m_b;
- double m_Lambda;
+ double m_Lambda; // regularization factor
+
+ vnl_vector<double> row_sums; // number of active weights per row
+ vnl_vector<double> local_weight_means; // mean weight of each row
void SetProblem(vnl_sparse_matrix< double >& A, vnl_vector<double>& b, double lambda)
{
S = new vnl_sparse_matrix_linear_system<double>(A, b);
m_A = A;
m_b = b;
m_Lambda = lambda;
+
+ m_A_Ones.set_size(m_A.rows(), m_A.cols());
+ m_A.reset();
+ while (m_A.next())
+ m_A_Ones.put(m_A.getrow(), m_A.getcolumn(), 1);
+
+ unsigned int N = m_b.size();
+ vnl_vector<double> ones; ones.set_size(dim); ones.fill(1.0);
+ row_sums.set_size(N);
+ m_A_Ones.mult(ones, row_sums);
+ local_weight_means.set_size(N);
}
VnlCostFunction(const int NumVars) : vnl_cost_function(NumVars)
{
}
- double f_itk(vnl_vector<double> const &x) const
+ void regu_MSE(vnl_vector<double> const &x, double& cost)
{
- double min = x.min_value();
- if( min<0 )
- return 10000 * -min;
+ double mean = x.mean();
+ vnl_vector<double> tx = x-mean;
+ cost += m_Lambda*1e8*tx.squared_magnitude()/x.size();
+ }
- double cost = S->get_rms_error(x);
- double regu = m_Lambda*1e8*x.squared_magnitude()/x.size();
- cost += regu;
- return cost;
+ void regu_MSM(vnl_vector<double> const &x, double& cost)
+ {
+ cost += m_Lambda*1e8*x.squared_magnitude()/x.size();
}
- double f(vnl_vector<double> const &x)
+ void regu_localMSE(vnl_vector<double> const &x, double& cost)
{
- // // cost for x (mean squared error)
- double cost = S->get_rms_error(x);
- cost *= cost;
+ m_A_Ones.mult(x, local_weight_means);
+ local_weight_means = element_quotient(local_weight_means, row_sums);
-// double cost = 0;
-// vnl_vector<double> d; d.set_size(m_b.size());
-// S->multiply(x, d);
-// d -= m_b;
-// for (int i=0; i<d.size(); i++)
-// {
+ m_A_Ones.reset();
+ unsigned int num_elements = 0;
+ double regu = 0;
+ while (m_A_Ones.next())
+ {
+ double d = 0;
+ if (x[m_A_Ones.getcolumn()]>local_weight_means[m_A_Ones.getrow()])
+ d = std::exp(x[m_A_Ones.getcolumn()]) - std::exp(local_weight_means[m_A_Ones.getrow()]);
+ else
+ d = x[m_A_Ones.getcolumn()] - local_weight_means[m_A_Ones.getrow()];
+ regu += d*d;
+ ++num_elements;
+ }
+ cost += m_Lambda*regu/num_elements;
+ }
+
+ void grad_regu_MSE(vnl_vector<double> const &x, vnl_vector<double> &dx)
+ {
+ double mean = x.mean();
+ vnl_vector<double> tx = x-mean;
+
+ vnl_vector<double> tx2(dim, 0.0);
+ vnl_vector<double> h(dim, 1.0);
+ for (int c=0; c<dim; c++)
+ {
+ h[c] = dim-1;
+ tx2[c] += dot_product(h,tx);
+ h[c] = 1;
+ }
+ dx += tx2*m_Lambda*1e8*2.0/(dim*dim);
+
+ }
+
+ void grad_regu_MSM(vnl_vector<double> const &x, vnl_vector<double> &dx)
+ {
+ dx += m_Lambda*1e8*2.0*x/dim;
+ }
+ void grad_regu_localMSE(vnl_vector<double> const &x, vnl_vector<double> &dx)
+ {
+ m_A_Ones.mult(x, local_weight_means);
+ local_weight_means = element_quotient(local_weight_means, row_sums);
+
+ vnl_vector<double> exp_x = x.apply(std::exp);
+ vnl_vector<double> exp_means = local_weight_means.apply(std::exp);
+
+ vnl_vector<double> tdx(dim, 0);
+ m_A_Ones.reset();
+ while (m_A_Ones.next())
+ {
+ int c = m_A_Ones.getcolumn();
+ int r = m_A_Ones.getrow();
+ if (row_sums[r]==0)
+ continue;
+
+ if (x[c]>local_weight_means[r])
+ tdx[c] += (exp_x[c] * ( exp_x[c] - exp_means[r] ))/row_sums[r];
+ else
+ tdx[c] += (x[c] - local_weight_means[r])/row_sums[r];
+ }
+ dx += tdx*2.0*m_Lambda;
+
+// vnl_vector<double> dr; dr.set_size(dim); dr.fill(0);
+// for (unsigned int r=0; r<m_A_Ones.rows(); r++)
+// {
+// int n = row_sums[r];
+// vnl_vector<double> weights; weights.set_size(n);
+// vnl_matrix<double> temp(n,n,1); temp.fill_diagonal(n-1);
+
+// int i=0;
+// for (auto w : m_A_Ones.get_row(r))
+// {
+// weights[i]=w.second;
+// ++i;
+// }
+
+// weights -= local_weight_means[r];
+// weights = temp*weights;
+
+// i=0;
+// for (auto w : m_A_Ones.get_row(r))
+// {
+// dr[w.second] += weights[i];
+// ++i;
+// }
// }
+// dx += dr*2.0*m_Lambda;
+ }
+
+
+ double f(vnl_vector<double> const &x)
+ {
+ double cost = S->get_rms_error(x);
+ cost *= cost;
+
// cost for e^x
// vnl_vector<double> x_exp; x_exp.set_size(x.size());
// for (unsigned int c=0; c<x.size(); c++)
// x_exp[c] = std::exp(x[c]);
- // vnl_vector<double> d; d.set_size(x.size());
- // S->multiply(x_exp, d);
- // d -= m_b;
- // double cost = d.squared_magnitude()/x.size();
-
- // // Tikhonov regu
- double regu = m_Lambda*1e8*x.squared_magnitude()/x.size();
-
- // aTV regu
- // double regu = 0;
- // unsigned int N = m_b.size();
- // vnl_vector<double> ones; ones.set_size(x.size()); ones.fill(1.0);
- // vnl_vector<double> means; means.set_size(N);
- // S->multiply(ones, means);
-
- // unsigned int norm = 0;
- // for (unsigned int i=0; i<N; ++i)
- // {
- // if (m_A.get_row(i).empty())
- // continue;
- // means[i] /= m_A.get_row(i).size();
+ // double cost = S->get_rms_error(x_exp);
+ // cost *= cost;
- // for (auto el : m_A.get_row(i))
- // {
- // float d = 0;
- // if (x[el.first]>means[i])
- // d = std::exp(x[el.first]) - std::exp(means[i]);
- // else
- // d = x[el.first] - means[i];
-
- // regu += d*d;
- // norm++;
- // }
- // }
- // regu /= norm;
+ regu_localMSE(x, cost);
+// regu_MSM(x, cost);
- cost += regu;
return cost;
}
void gradf(vnl_vector<double> const &x, vnl_vector<double> &dx)
{
dx.fill(0.0);
unsigned int N = m_b.size();
// vnl_vector<double> x_exp; x_exp.set_size(x.size());
// for (unsigned int c=0; c<x.size(); c++)
// x_exp[c] = std::exp(x[c]);
vnl_vector<double> d; d.set_size(N);
S->multiply(x,d);
d -= m_b;
S->transpose_multiply(d, dx);
dx *= 2.0/N;
// for (unsigned int c=0; c<x.size(); c++)
// dx[c] *= x_exp[c]; // only for e^x weights
- dx += m_Lambda*1e8*2.0*x/x.size();
+ grad_regu_localMSE(x,dx);
+// grad_regu_MSM(x,dx);
}
-
};
-class ItkCostFunction : public itk::SingleValuedCostFunction
-{
-public:
- /** Standard class typedefs. */
- typedef ItkCostFunction Self;
- typedef SingleValuedCostFunction Superclass;
- typedef itk::SmartPointer<Self> Pointer;
- typedef itk::SmartPointer<const Self> ConstPointer;
-
- /** Method for creation through the object factory. */
- itkNewMacro(Self)
-
- /** Run-time type information (and related methods). */
- itkTypeMacro(ItkCostFunction, SingleValuedCostfunction)
-
- void SetVnlCostFunction(VnlCostFunction& cf) { m_VnlCostFunction = cf; }
- unsigned int GetNumberOfParameters(void) const { return m_VnlCostFunction.get_number_of_unknowns(); } // itk::CostFunction
-
- MeasureType GetValue(const ParametersType & parameters) const
- {
- return m_VnlCostFunction.f_itk(parameters);
- }
-
- void GetDerivative(const ParametersType &, DerivativeType & ) const {
- throw itk::ExceptionObject( __FILE__, __LINE__, "No derivative is available for this cost function.");
- }
-
-protected:
- ItkCostFunction(){}
- ~ItkCostFunction(){}
-
- VnlCostFunction m_VnlCostFunction = VnlCostFunction(1);
-
-private:
- ItkCostFunction(const Self &); //purposely not implemented
- void operator = (const Self &); //purposely not implemented
-};
-
-void OptimizeItk(VnlCostFunction& cf, vnl_vector<double>& x, int iter, double lb, double ub)
-{
- ItkCostFunction::ParametersType p; p.SetData(x.data_block(), x.size());
-
- ItkCostFunction::Pointer itk_cf = ItkCostFunction::New();
- itk_cf->SetVnlCostFunction(cf);
-
- std::pair< double, double > bounds; bounds.first = lb; bounds.second = ub;
- MITK_INFO << bounds;
-
- // itk::Statistics::MersenneTwisterRandomVariateGenerator::Pointer randGen = itk::Statistics::MersenneTwisterRandomVariateGenerator::New();
- // itk::OnePlusOneEvolutionaryOptimizer::Pointer opt = itk::OnePlusOneEvolutionaryOptimizer::New();
- // opt->SetCostFunction(itk_cf);
- // opt->MinimizeOn();
- // opt->SetInitialPosition(p);
- // opt->SetNormalVariateGenerator(randGen);
- // opt->SetMaximumIteration(iter);
- // opt->StartOptimization();
-
- // itk::ParticleSwarmOptimizer::Pointer opt = itk::ParticleSwarmOptimizer::New();
- // opt->SetCostFunction(itk_cf);
- // opt->SetInitialPosition(p);
- // opt->SetParameterBounds(bounds, x.size());
- // opt->SetMaximalNumberOfIterations(iter);
- // opt->SetNumberOfParticles(100);
- // opt->SetParametersConvergenceTolerance(0.01, x.size());
- // opt->SetNumberOfGenerationsWithMinimalImprovement(3);
- // opt->StartOptimization();
-
- // itk::GradientDescentOptimizer::Pointer opt = itk::GradientDescentOptimizer::New();
- // opt->SetCostFunction(itk_cf);
- // opt->SetInitialPosition(p);
- // opt->SetMinimize(true);
- // opt->SetNumberOfIterations(iter);
- // opt->StartOptimization();
-
- itk::SPSAOptimizer::Pointer opt = itk::SPSAOptimizer::New();
- opt->SetCostFunction(itk_cf);
- opt->SetInitialPosition(p);
- opt->SetMinimize(true);
- opt->SetNumberOfPerturbations(iter);
- opt->StartOptimization();
-
- x.copy_in(opt->GetCurrentPosition().data_block());
- for (unsigned int i=0; i<x.size(); i++)
- MITK_INFO << opt->GetCurrentPosition()[i];
- // MITK_INFO << "Cost: " << opt->GetCurrentCost();
-
-}
-
-std::vector<float> FitFibers( std::string , std::vector< mitk::FiberBundle::Pointer > input_tracts, mitk::Image::Pointer inputImage, bool single_fiber_fit, int max_iter, float g_tol, float lambda )
+vnl_vector<double> FitFibers( std::string , std::vector< mitk::FiberBundle::Pointer > input_tracts, mitk::Image::Pointer inputImage, vnl_sparse_matrix< double >& A, vnl_vector<double>& b, bool single_fiber_fit, int max_iter, float g_tol, float lambda )
{
typedef mitk::ImageToItk< PeakImgType > CasterType;
CasterType::Pointer caster = CasterType::New();
caster->SetInput(inputImage);
caster->Update();
PeakImgType::Pointer itkImage = caster->GetOutput();
unsigned int* image_size = inputImage->GetDimensions();
int sz_x = image_size[0];
int sz_y = image_size[1];
int sz_z = image_size[2];
int sz_peaks = image_size[3]/3 + 1; // +1 for zero - peak
int num_voxels = sz_x*sz_y*sz_z;
unsigned int num_unknowns = input_tracts.size();
if (single_fiber_fit)
{
num_unknowns = 0;
for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
num_unknowns += input_tracts.at(bundle)->GetNumFibers();
}
unsigned int number_of_residuals = num_voxels * sz_peaks;
// create linear system
MITK_INFO << "Num. unknowns: " << num_unknowns;
MITK_INFO << "Num. residuals: " << number_of_residuals;
MITK_INFO << "Creating system ...";
- vnl_sparse_matrix< double > A; A.set_size(number_of_residuals, num_unknowns);
- vnl_vector<double> b; b.set_size(number_of_residuals); b.fill(0.0);
+ A.set_size(number_of_residuals, num_unknowns);
+ b.set_size(number_of_residuals); b.fill(0.0);
double TD = 0;
double FD = 0;
unsigned int dir_count = 0;
unsigned int fiber_count = 0;
for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
{
vtkSmartPointer<vtkPolyData> polydata = input_tracts.at(bundle)->GetFiberPolyData();
for (int i=0; i<input_tracts.at(bundle)->GetNumFibers(); ++i)
{
vtkCell* cell = polydata->GetCell(i);
int numPoints = cell->GetNumberOfPoints();
vtkPoints* points = cell->GetPoints();
if (numPoints<2)
MITK_INFO << "FIBER WITH ONLY ONE POINT ENCOUNTERED!";
for (int j=0; j<numPoints-1; ++j)
{
double* p1 = points->GetPoint(j);
PointType4 p;
p[0]=p1[0];
p[1]=p1[1];
p[2]=p1[2];
p[3]=0;
itk::Index<4> idx4;
itkImage->TransformPhysicalPointToIndex(p, idx4);
if (!itkImage->GetLargestPossibleRegion().IsInside(idx4))
continue;
double* p2 = points->GetPoint(j+1);
vnl_vector_fixed<float,3> fiber_dir;
fiber_dir[0] = p[0]-p2[0];
fiber_dir[1] = p[1]-p2[1];
fiber_dir[2] = p[2]-p2[2];
fiber_dir.normalize();
double w = 1;
int peak_id = sz_peaks-1;
vnl_vector_fixed<float,3> odf_peak = GetClosestPeak(idx4, itkImage, fiber_dir, peak_id, w);
float peak_mag = odf_peak.magnitude();
int x = idx4[0];
int y = idx4[1];
int z = idx4[2];
- unsigned int linear_index = sz_peaks*(x + sz_x*y + sz_x*sz_y*z);
+ unsigned int linear_index = x + sz_x*y + sz_x*sz_y*z + sz_x*sz_y*sz_z*peak_id;
- if (b[linear_index + peak_id] == 0 && peak_id<3)
+ if (b[linear_index] == 0 && peak_id<3)
{
dir_count++;
FD += peak_mag;
}
TD += w;
if (single_fiber_fit)
{
- b[linear_index + peak_id] = peak_mag;
- A.put(linear_index + peak_id, fiber_count, A.get(linear_index + peak_id, fiber_count) + w);
+ b[linear_index] = peak_mag;
+ A.put(linear_index, fiber_count, A.get(linear_index, fiber_count) + w);
}
else
{
- b[linear_index + peak_id] = peak_mag;
- A.put(linear_index + peak_id, bundle, A.get(linear_index + peak_id, bundle) + w);
+ b[linear_index] = peak_mag;
+ A.put(linear_index, bundle, A.get(linear_index, bundle) + w);
}
}
++fiber_count;
}
}
TD /= (dir_count*fiber_count);
FD /= dir_count;
- A *= 1.0/TD;
- b *= 100.0*UPSCALE/FD; // times 100 because we want to avoid too small weights
+ A /= TD;
+ b *= 100.0/FD; // times 100 because we want to avoid too small weights
MITK_INFO << "TD: " << TD;
MITK_INFO << "FD: " << FD;
MITK_INFO << "Regularization: " << lambda; // pretty large regularization needed --> internally upscale by 1e8
itk::TimeProbe clock;
clock.Start();
MITK_INFO << "Fitting fibers";
VnlCostFunction cost(num_unknowns);
cost.SetProblem(A, b, lambda);
vnl_vector<double> x; x.set_size(num_unknowns); x.fill( TD/100.0 * FD/2.0 );
vnl_lbfgsb minimizer(cost);
vnl_vector<double> l; l.set_size(num_unknowns); l.fill(0);
vnl_vector<long> bound_selection; bound_selection.set_size(num_unknowns); bound_selection.fill(1);
minimizer.set_bound_selection(bound_selection);
minimizer.set_lower_bound(l);
minimizer.set_trace(true);
minimizer.set_projected_gradient_tolerance(g_tol);
if (max_iter>0)
minimizer.set_max_function_evals(max_iter);
minimizer.minimize(x);
// SECOND RUN // USE QUARTILE AS LIMIT???
float mean_w = x.mean();
- x.fill(mean_w);
MITK_INFO << "Mean weight: " << mean_w;
+ x.fill(mean_w);
vnl_vector<double> u; u.set_size(num_unknowns); u.fill(mean_w * 3);
minimizer.set_upper_bound(u);
bound_selection.fill(2);
minimizer.set_bound_selection(bound_selection);
minimizer.minimize(x);
MITK_INFO << "Residual error: " << minimizer.get_end_error();
MITK_INFO << "NumEvals: " << minimizer.get_num_evaluations();
MITK_INFO << "NumIterations: " << minimizer.get_num_iterations();
clock.Stop();
int h = clock.GetTotal()/3600;
int m = ((int)clock.GetTotal()%3600)/60;
int s = (int)clock.GetTotal()%60;
MITK_INFO << "Optimization took " << h << "h, " << m << "m and " << s << "s";
- std::vector<float> weights;
- for (unsigned int i=0; i<num_unknowns; ++i)
- {
-// MITK_INFO << x[i];
- weights.push_back(FD*x[i]);
-// weights.push_back(x[i]);
- }
- return weights;
+// vnl_sparse_matrix_linear_system<double> S(A, b);
+// vnl_vector<double> fitted_b; fitted_b.set_size(b.size());
+// S.multiply(x, fitted_b);
+// for (unsigned int r=0; r<b.size(); r++)
+// MITK_INFO << fitted_b[r] << " - " << b[r];
+
+ return x;
}
std::vector< string > get_file_list(const std::string& path)
{
std::vector< string > file_list;
itk::Directory::Pointer dir = itk::Directory::New();
if (dir->Load(path.c_str()))
{
int n = dir->GetNumberOfFiles();
for (int r = 0; r < n; r++)
{
const char *filename = dir->GetFile(r);
std::string ext = ist::GetFilenameExtension(filename);
if (ext==".fib" || ext==".trk")
file_list.push_back(path + '/' + filename);
}
}
return file_list;
}
int main(int argc, char* argv[])
{
mitkCommandLineParser parser;
parser.setTitle("Fit Fibers To Image");
parser.setCategory("Fiber Tracking Evaluation");
parser.setDescription("");
parser.setContributor("MIC");
parser.setArgumentPrefix("--", "-");
parser.addArgument("", "i1", mitkCommandLineParser::StringList, "Input tractograms:", "input tractograms (.fib, vtk ascii file format)", us::Any(), false);
parser.addArgument("", "i2", mitkCommandLineParser::InputFile, "Input peaks:", "input peak image", us::Any(), false);
parser.addArgument("", "it", mitkCommandLineParser::Int, "", "");
- parser.addArgument("", "s", mitkCommandLineParser::Bool, "", "");
+ parser.addArgument("", "b", mitkCommandLineParser::Bool, "", "");
parser.addArgument("", "g", mitkCommandLineParser::Float, "", "");
parser.addArgument("", "l", mitkCommandLineParser::Float, "", "");
+ parser.addArgument("", "r", mitkCommandLineParser::Int, "", "");
parser.addArgument("", "o", mitkCommandLineParser::OutputDirectory, "Output:", "output root", us::Any(), false);
map<string, us::Any> parsedArgs = parser.parseArguments(argc, argv);
if (parsedArgs.size()==0)
return EXIT_FAILURE;
mitkCommandLineParser::StringContainerType fib_files = us::any_cast<mitkCommandLineParser::StringContainerType>(parsedArgs["i1"]);
string dwiFile = us::any_cast<string>(parsedArgs["i2"]);
string outRoot = us::any_cast<string>(parsedArgs["o"]);
- bool single_fib = false;
- if (parsedArgs.count("s"))
- single_fib = us::any_cast<bool>(parsedArgs["s"]);
+ bool single_fib = true;
+ if (parsedArgs.count("b"))
+ single_fib = !us::any_cast<bool>(parsedArgs["b"]);
int max_iter = 0;
if (parsedArgs.count("it"))
max_iter = us::any_cast<int>(parsedArgs["it"]);
+// int regu = 0;
+// if (parsedArgs.count("r"))
+// regu = us::any_cast<int>(parsedArgs["r"]);
+
float g_tol = 1e-5;
if (parsedArgs.count("g"))
g_tol = us::any_cast<float>(parsedArgs["g"]);
float lambda = 1.0;
if (parsedArgs.count("l"))
lambda = us::any_cast<float>(parsedArgs["l"]);
try
{
std::vector< mitk::FiberBundle::Pointer > input_tracts;
mitk::PreferenceListReaderOptionsFunctor functor = mitk::PreferenceListReaderOptionsFunctor({"Peak Image", "Fiberbundles"}, {});
mitk::Image::Pointer inputImage = dynamic_cast<mitk::PeakImage*>(mitk::IOUtil::Load(dwiFile, &functor)[0].GetPointer());
float minSpacing = 1;
if(inputImage->GetGeometry()->GetSpacing()[0]<inputImage->GetGeometry()->GetSpacing()[1] && inputImage->GetGeometry()->GetSpacing()[0]<inputImage->GetGeometry()->GetSpacing()[2])
minSpacing = inputImage->GetGeometry()->GetSpacing()[0];
else if (inputImage->GetGeometry()->GetSpacing()[1] < inputImage->GetGeometry()->GetSpacing()[2])
minSpacing = inputImage->GetGeometry()->GetSpacing()[1];
else
minSpacing = inputImage->GetGeometry()->GetSpacing()[2];
std::vector< std::string > fib_names;
for (auto item : fib_files)
{
if ( ist::FileIsDirectory(item) )
{
for ( auto fibFile : get_file_list(item) )
{
mitk::FiberBundle::Pointer inputTractogram = dynamic_cast<mitk::FiberBundle*>(mitk::IOUtil::Load(fibFile)[0].GetPointer());
if (inputTractogram.IsNull())
continue;
- inputTractogram->ResampleLinear(minSpacing/4);
+ inputTractogram->ResampleLinear(minSpacing/10);
input_tracts.push_back(inputTractogram);
fib_names.push_back(fibFile);
}
}
else
{
mitk::FiberBundle::Pointer inputTractogram = dynamic_cast<mitk::FiberBundle*>(mitk::IOUtil::Load(item)[0].GetPointer());
if (inputTractogram.IsNull())
continue;
- inputTractogram->ResampleLinear(minSpacing/4);
+ inputTractogram->ResampleLinear(minSpacing/10);
input_tracts.push_back(inputTractogram);
fib_names.push_back(item);
}
}
- std::vector<float> weights = FitFibers(outRoot, input_tracts, inputImage, single_fib, max_iter, g_tol, lambda);
+ vnl_sparse_matrix<double> A;
+ vnl_vector<double> b;
+ vnl_vector<double> x = FitFibers(outRoot, input_tracts, inputImage, A, b, single_fib, max_iter, g_tol, lambda);
+ MITK_INFO << "Weighting fibers";
if (single_fib)
{
unsigned int fiber_count = 0;
for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
{
for (int i=0; i<input_tracts.at(bundle)->GetNumFibers(); i++)
{
- input_tracts.at(bundle)->SetFiberWeight(i, weights.at(fiber_count));
+ input_tracts.at(bundle)->SetFiberWeight(i, x[fiber_count]);
++fiber_count;
}
}
}
else
{
for (unsigned int i=0; i<fib_names.size(); ++i)
- input_tracts.at(i)->SetFiberWeights(weights.at(i));
+ input_tracts.at(i)->SetFiberWeights(x[i]);
}
// OUTPUT IMAGES
MITK_INFO << "Generating output images ...";
typedef mitk::ImageToItk< PeakImgType > CasterType;
CasterType::Pointer caster = CasterType::New();
caster->SetInput(inputImage);
caster->Update();
- PeakImgType::Pointer itkImage = caster->GetOutput();
+ PeakImgType::Pointer peak_image = caster->GetOutput();
itk::ImageDuplicator< PeakImgType >::Pointer duplicator = itk::ImageDuplicator< PeakImgType >::New();
- duplicator->SetInputImage(itkImage);
+ duplicator->SetInputImage(peak_image);
duplicator->Update();
PeakImgType::Pointer unexplained_image = duplicator->GetOutput();
+ unexplained_image->FillBuffer(0.0);
duplicator->SetInputImage(unexplained_image);
duplicator->Update();
PeakImgType::Pointer residual_image = duplicator->GetOutput();
+ residual_image->FillBuffer(0.0);
duplicator->SetInputImage(residual_image);
duplicator->Update();
- PeakImgType::Pointer explained_image = duplicator->GetOutput();
- explained_image->FillBuffer(0.0);
-
- for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
- {
- vtkSmartPointer<vtkPolyData> polydata = input_tracts.at(bundle)->GetFiberPolyData();
-
- for (int i=0; i<input_tracts.at(bundle)->GetNumFibers(); ++i)
- {
- vtkCell* cell = polydata->GetCell(i);
- int numPoints = cell->GetNumberOfPoints();
- vtkPoints* points = cell->GetPoints();
-
- if (numPoints<2)
- MITK_INFO << "FIBER WITH ONLY ONE POINT ENCOUNTERED!";
-
- float w = input_tracts.at(bundle)->GetFiberWeight(i);
-
- for (int j=0; j<numPoints-1; ++j)
- {
- double* p1 = points->GetPoint(j);
- PointType4 p;
- p[0]=p1[0];
- p[1]=p1[1];
- p[2]=p1[2];
- p[3]=0;
-
- itk::Index<4> idx4;
- itkImage->TransformPhysicalPointToIndex(p, idx4);
- if (!itkImage->GetLargestPossibleRegion().IsInside(idx4))
- continue;
-
- double* p2 = points->GetPoint(j+1);
- vnl_vector_fixed<float,3> fiber_dir;
- fiber_dir[0] = p[0]-p2[0];
- fiber_dir[1] = p[1]-p2[1];
- fiber_dir[2] = p[2]-p2[2];
- fiber_dir.normalize();
-
- int peak_id = -1;
- double contr = 1;
- GetClosestPeak(idx4, itkImage, fiber_dir, peak_id, contr);
- if (peak_id<0)
- continue;
-
- vnl_vector_fixed<float,3> unexplained_dir;
- vnl_vector_fixed<float,3> explained_dir;
- vnl_vector_fixed<float,3> res_dir;
- vnl_vector_fixed<float,3> peak_dir;
-
- idx4[3] = peak_id*3;
- unexplained_dir[0] = unexplained_image->GetPixel(idx4);
- explained_dir[0] = explained_image->GetPixel(idx4);
- res_dir[0] = residual_image->GetPixel(idx4);
- peak_dir[0] = itkImage->GetPixel(idx4);
-
- idx4[3] += 1;
- unexplained_dir[1] = unexplained_image->GetPixel(idx4);
- explained_dir[1] = explained_image->GetPixel(idx4);
- res_dir[1] = residual_image->GetPixel(idx4);
- peak_dir[1] = itkImage->GetPixel(idx4);
-
- idx4[3] += 1;
- unexplained_dir[2] = unexplained_image->GetPixel(idx4);
- explained_dir[2] = explained_image->GetPixel(idx4);
- res_dir[2] = residual_image->GetPixel(idx4);
- peak_dir[2] = itkImage->GetPixel(idx4);
-
- if (dot_product(peak_dir, fiber_dir)<0)
- fiber_dir *= -1;
- fiber_dir *= w;
-
- idx4[3] = peak_id*3;
- residual_image->SetPixel(idx4, res_dir[0] - fiber_dir[0]);
-
- idx4[3] += 1;
- residual_image->SetPixel(idx4, res_dir[1] - fiber_dir[1]);
-
- idx4[3] += 1;
- residual_image->SetPixel(idx4, res_dir[2] - fiber_dir[2]);
-
- if ( fabs(unexplained_dir[0]) - fabs(fiber_dir[0]) < 0 ) // did we "overexplain" stuff?
- fiber_dir = unexplained_dir;
+ PeakImgType::Pointer fitted_image = duplicator->GetOutput();
+ fitted_image->FillBuffer(0.0);
+
+// vnl_sparse_matrix_linear_system<double> S(A, b);
+// vnl_vector<double> fitted_b; fitted_b.set_size(b.size());
+// S.multiply(x, fitted_b);
+
+// unsigned int* image_size = inputImage->GetDimensions();
+// int sz_x = image_size[0];
+// int sz_y = image_size[1];
+// int sz_z = image_size[2];
+// int sz_peaks = image_size[3]/3 + 1; // +1 for zero - peak
+// for (unsigned int r=0; r<b.size(); r++)
+// {
+// itk::Index<4> idx;
+// unsigned int linear_index = r;
+// idx[0] = linear_index % sz_x; linear_index /= sz_x;
+// idx[1] = linear_index % sz_y; linear_index /= sz_y;
+// idx[2] = linear_index % sz_z; linear_index /= sz_z;
+// int peak_id = linear_index % sz_peaks;
+
+// if (peak_id<sz_peaks-1)
+// {
+// vnl_vector_fixed<float,3> peak_dir;
+
+// idx[3] = peak_id*3;
+// peak_dir[0] = peak_image->GetPixel(idx);
+// idx[3] += 1;
+// peak_dir[1] = peak_image->GetPixel(idx);
+// idx[3] += 1;
+// peak_dir[2] = peak_image->GetPixel(idx);
+
+// peak_dir.normalize();
+// peak_dir *= fitted_b[r];
+// // MITK_INFO << fitted_b[r] << " - " << b[r];
+
+// idx[3] = peak_id*3;
+// fitted_image->SetPixel(idx, peak_dir[0]);
+
+// idx[3] += 1;
+// fitted_image->SetPixel(idx, peak_dir[1]);
+
+// idx[3] += 1;
+// fitted_image->SetPixel(idx, peak_dir[2]);
+// }
+// }
- idx4[3] = peak_id*3;
- unexplained_image->SetPixel(idx4, unexplained_dir[0] - fiber_dir[0]);
- explained_image->SetPixel(idx4, explained_dir[0] + fiber_dir[0]);
- idx4[3] += 1;
- unexplained_image->SetPixel(idx4, unexplained_dir[1] - fiber_dir[1]);
- explained_image->SetPixel(idx4, explained_dir[1] + fiber_dir[1]);
+ // for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
+ // {
+ // vtkSmartPointer<vtkPolyData> polydata = input_tracts.at(bundle)->GetFiberPolyData();
- idx4[3] += 1;
- unexplained_image->SetPixel(idx4, unexplained_dir[2] - fiber_dir[2]);
- explained_image->SetPixel(idx4, explained_dir[2] + fiber_dir[2]);
- }
+ // for (int i=0; i<input_tracts.at(bundle)->GetNumFibers(); ++i)
+ // {
+ // vtkCell* cell = polydata->GetCell(i);
+ // int numPoints = cell->GetNumberOfPoints();
+ // vtkPoints* points = cell->GetPoints();
+
+ // if (numPoints<2)
+ // MITK_INFO << "FIBER WITH ONLY ONE POINT ENCOUNTERED!";
+
+ // float w = input_tracts.at(bundle)->GetFiberWeight(i);
+
+ // for (int j=0; j<numPoints-1; ++j)
+ // {
+ // double* p1 = points->GetPoint(j);
+ // PointType4 p;
+ // p[0]=p1[0];
+ // p[1]=p1[1];
+ // p[2]=p1[2];
+ // p[3]=0;
+
+ // itk::Index<4> idx4;
+ // peak_image->TransformPhysicalPointToIndex(p, idx4);
+ // if (!peak_image->GetLargestPossibleRegion().IsInside(idx4))
+ // continue;
+
+ // double* p2 = points->GetPoint(j+1);
+ // vnl_vector_fixed<float,3> tract_dir;
+ // tract_dir[0] = p[0]-p2[0];
+ // tract_dir[1] = p[1]-p2[1];
+ // tract_dir[2] = p[2]-p2[2];
+ // tract_dir.normalize();
+
+ // int peak_id = -1;
+ // double contr = 1;
+ // vnl_vector_fixed<float,3> peak_dir = GetClosestPeak(idx4, peak_image, tract_dir, peak_id, contr);
+
+ // idx4[3] = peak_id*3;
+ // tract_image->SetPixel(idx4, tract_image->GetPixel(idx4) + tract_dir[0]);
+
+ // idx4[3] += 1;
+ // tract_image->SetPixel(idx4, tract_image->GetPixel(idx4) + tract_dir[1]);
+
+ // idx4[3] += 1;
+ // tract_image->SetPixel(idx4, tract_image->GetPixel(idx4) + tract_dir[2]);
+ // }
- }
- }
+ // }
+ // }
- itk::ImageFileWriter< PeakImgType >::Pointer writer = itk::ImageFileWriter< PeakImgType >::New();
- writer->SetInput(unexplained_image);
- writer->SetFileName(outRoot + "unexplained_image.nrrd");
- writer->Update();
+ // itk::ImageFileWriter< PeakImgType >::Pointer writer = itk::ImageFileWriter< PeakImgType >::New();
+ // writer->SetInput(fitted_image);
+ // writer->SetFileName(outRoot + "fitted_image.nrrd");
+ // writer->Update();
- writer->SetInput(explained_image);
- writer->SetFileName(outRoot + "explained_image.nrrd");
- writer->Update();
+ // writer->SetInput(tract_image);
+ // writer->SetFileName(outRoot + "explained_image.nrrd");
+ // writer->Update();
- writer->SetInput(residual_image);
- writer->SetFileName(outRoot + "residual_image.nrrd");
- writer->Update();
+ // writer->SetInput(residual_image);
+ // writer->SetFileName(outRoot + "residual_image.nrrd");
+ // writer->Update();
for (unsigned int bundle=0; bundle<input_tracts.size(); bundle++)
{
input_tracts.at(bundle)->Compress(0.1);
std::string name = fib_names.at(bundle);
name = ist::GetFilenameWithoutExtension(name);
mitk::IOUtil::Save(input_tracts.at(bundle), outRoot + name + "_fitted.fib");
}
}
catch (itk::ExceptionObject e)
{
std::cout << e;
return EXIT_FAILURE;
}
catch (std::exception e)
{
std::cout << e.what();
return EXIT_FAILURE;
}
catch (...)
{
std::cout << "ERROR!?!";
return EXIT_FAILURE;
}
return EXIT_SUCCESS;
}