bool output_odfs(const image::basic_image<unsigned char,3>& mni_mask,
const char* out_name,
const char* ext,
std::vector<std::vector<float> >& odfs,
const tessellated_icosahedron& ti,
const float* vs,
const float* mni,
const std::string& report,
bool record_odf = true)
{
begin_prog("output");
ImageModel image_model;
if(report.length())
image_model.voxel.report = report.c_str();
image_model.voxel.dim = mni_mask.geometry();
image_model.voxel.ti = ti;
image_model.voxel.odf_decomposition = false;
image_model.voxel.odf_deconvolusion = false;
image_model.voxel.half_sphere = false;
image_model.voxel.max_fiber_number = 5;
image_model.voxel.z0 = 0.0;
image_model.voxel.need_odf = record_odf;
image_model.voxel.template_odfs.swap(odfs);
image_model.voxel.param = mni;
image_model.file_name = out_name;
image_model.mask = mni_mask;
std::copy(vs,vs+3,image_model.voxel.vs.begin());
if (prog_aborted() || !image_model.reconstruct<reprocess_odf>(1))
return false;
image_model.save_fib(ext);
image_model.voxel.template_odfs.swap(odfs);
return true;
}
void trim(void)
{
image::geometry<3> range_min,range_max;
image::bounding_box(mask,range_min,range_max,0);
for (unsigned int index = 0;check_prog(index,dwi_data.size());++index)
{
image::pointer_image<unsigned short,3> I = image::make_image(voxel.dim,(unsigned short*)dwi_data[index]);
image::basic_image<unsigned short,3> I0 = I;
image::crop(I0,range_min,range_max);
std::fill(I.begin(),I.end(),0);
std::copy(I0.begin(),I0.end(),I.begin());
}
calculate_dwi_sum();
image::crop(mask,range_min,range_max);
voxel.dim = mask.geometry();
}
virtual void init(Voxel& voxel)
{
if(voxel.vs[0] == 0.0 ||
voxel.vs[1] == 0.0 ||
voxel.vs[2] == 0.0)
throw std::runtime_error("No spatial information found in src file. Recreate src file or contact developer for assistance");
begin_prog("normalization");
VG = fa_template_imp.I;
VF = voxel.fa_map;
image::filter::gaussian(voxel.fa_map);
image::filter::gaussian(voxel.fa_map);
image::filter::gaussian(voxel.qa_map);
image::filter::gaussian(voxel.qa_map);
image::normalize(voxel.fa_map,1.0);
image::normalize(voxel.qa_map,1.0);
for(unsigned int index = 0;index < voxel.qa_map.size();++index)
if(voxel.qa_map[index] == 0.0 || voxel.fa_map[index]/voxel.qa_map[index] > 2.5)
VF[index] = 0.0;
image::filter::gaussian(VF);
src_geo = VF.geometry();
image::normalize(VF,1.0);
//VF.save_to_file<image::io::nifti<> >("VF.nii");
image::normalize(VG,1.0);
// get rid of the gray matters
image::minus_constant(VF.begin(),VF.end(),0.3);
image::lower_threshold(VF.begin(),VF.end(),0.00);
image::normalize(VF,1.0);
image::minus_constant(VG.begin(),VG.end(),0.3);
image::lower_threshold(VG.begin(),VG.end(),0.00);
image::normalize(VG,1.0);
VGvs[0] = std::fabs(fa_template_imp.tran[0]);
VGvs[1] = std::fabs(fa_template_imp.tran[5]);
VGvs[2] = std::fabs(fa_template_imp.tran[10]);
image::affine_transform<3,double> arg_min;
// VG: FA TEMPLATE
// VF: SUBJECT QA
arg_min.scaling[0] = voxel.vs[0] / VGvs[0];
arg_min.scaling[1] = voxel.vs[1] / VGvs[1];
arg_min.scaling[2] = voxel.vs[2] / VGvs[2];
voxel_volume_scale = arg_min.scaling[0]*arg_min.scaling[1]*arg_min.scaling[2];
// calculate center of mass
image::vector<3,double> mF = center_of_mass(VF);
image::vector<3,double> mG = center_of_mass(VG);
arg_min.translocation[0] = mG[0]-mF[0]*arg_min.scaling[0];
arg_min.translocation[1] = mG[1]-mF[1]*arg_min.scaling[1];
arg_min.translocation[2] = mG[2]-mF[2]*arg_min.scaling[2];
bool terminated = false;
set_title("linear registration");
begin_prog("conducting registration");
check_prog(0,2);
image::reg::linear(VF,VG,arg_min,image::reg::affine,image::reg::square_error(),terminated,0.25);
check_prog(1,2);
// create VFF the affine transformed VF
image::basic_image<float,3> VFF(VG.geometry());
{
affine = arg_min;
image::reg::linear_get_trans(VF.geometry(),VG.geometry(),affine);
{
std::vector<double> T(16);
affine.save_to_transform(T.begin());
T[15] = 1.0;
math::matrix_inverse(T.begin(),math::dim<4,4>());
affine.load_from_transform(T.begin());
}
image::resample(VF,VFF,affine);
//VFF.save_to_file<image::io::nifti<> >("VFF.nii");
//VG.save_to_file<image::io::nifti<> >("VG.nii");
}
{
switch(voxel.reg_method)
{
case 0:
mni.normalize(VG,VFF);
break;
case 1:
mni.normalize(VG,VFF,12,14,12,4,8);
break;
}
//calculate the goodness of fit
std::vector<float> x,y;
x.reserve(VG.size());
y.reserve(VG.size());
image::interpolation<image::linear_weighting,3> trilinear_interpolation;
for(image::pixel_index<3> index;VG.geometry().is_valid(index);
index.next(VG.geometry()))
if(VG[index.index()] != 0)
{
image::vector<3,double> pos;
mni.warp_coordinate(index,pos);
double value = 0.0;
if(!trilinear_interpolation.estimate(VFF,pos,value))
continue;
x.push_back(VG[index.index()]);
y.push_back(value);
}
R2 = x.empty() ? 0.0 : image::correlation(x.begin(),x.end(),y.begin());
R2 *= R2;
std::cout << "R2 = " << R2 << std::endl;
}
check_prog(2,2);
// setup output bounding box
{
//setBoundingBox(-78,-112,-50,78,76,85,1.0);
float voxel_size = voxel.param[1];
bounding_box_lower[0] = std::floor(-78.0/voxel_size+0.5)*voxel_size;
bounding_box_lower[1] = std::floor(-112.0/voxel_size+0.5)*voxel_size;
bounding_box_lower[2] = std::floor(-50.0/voxel_size+0.5)*voxel_size;
bounding_box_upper[0] = std::floor(78.0/voxel_size+0.5)*voxel_size;
bounding_box_upper[1] = std::floor(76.0/voxel_size+0.5)*voxel_size;
bounding_box_upper[2] = std::floor(85.0/voxel_size+0.5)*voxel_size;
des_geo[0] = (bounding_box_upper[0]-bounding_box_lower[0])/voxel_size+1;//79
des_geo[1] = (bounding_box_upper[1]-bounding_box_lower[1])/voxel_size+1;//95
des_geo[2] = (bounding_box_upper[2]-bounding_box_lower[2])/voxel_size+1;//69
des_offset[0] = bounding_box_lower[0]/VGvs[0]-fa_template_imp.tran[3]/fa_template_imp.tran[0];
des_offset[1] = bounding_box_lower[1]/VGvs[1]-fa_template_imp.tran[7]/fa_template_imp.tran[5];
des_offset[2] = bounding_box_lower[2]/VGvs[2]-fa_template_imp.tran[11]/fa_template_imp.tran[10];
scale[0] = voxel_size/VGvs[0];
scale[1] = voxel_size/VGvs[1];
scale[2] = voxel_size/VGvs[2];
}
begin_prog("q-space diffeomorphic reconstruction");
float sigma = voxel.param[0]; //diffusion sampling length ratio, optimal 1.24
// setup mask
{
// set the current mask to template space
voxel.image_model->set_dimension(des_geo[0],des_geo[1],des_geo[2]);
for(image::pixel_index<3> index;des_geo.is_valid(index);index.next(des_geo))
{
image::vector<3,int> mni_pos(index);
mni_pos *= voxel.param[1];
mni_pos[0] /= VGvs[0];
mni_pos[1] /= VGvs[1];
mni_pos[2] /= VGvs[2];
mni_pos += des_offset;
voxel.image_model->mask[index.index()] =
fa_template_imp.I.at(mni_pos[0],mni_pos[1],mni_pos[2]) > 0.0? 1: 0;
}
}
q_vectors_time.resize(voxel.bvalues.size());
for (unsigned int index = 0; index < voxel.bvalues.size(); ++index)
{
q_vectors_time[index] = voxel.bvectors[index];
q_vectors_time[index] *= std::sqrt(voxel.bvalues[index]*0.01506);// get q in (mm) -1
q_vectors_time[index] *= sigma;
}
b0_index = -1;
if(voxel.half_sphere)
for(unsigned int index = 0;index < voxel.bvalues.size();++index)
if(voxel.bvalues[index] == 0)
b0_index = index;
ptr_images.clear();
for (unsigned int index = 0; index < voxel.image_model->dwi_data.size(); ++index)
ptr_images.push_back(point_image_type((const unsigned short*)voxel.image_model->dwi_data[index],src_geo));
voxel.qa_scaling = voxel.reponse_function_scaling/voxel.vs[0]/voxel.vs[1]/voxel.vs[2];
max_accumulated_qa = 0;
std::fill(voxel.vs.begin(),voxel.vs.end(),voxel.param[1]);
jdet.resize(des_geo.size());
if (voxel.odf_deconvolusion)
{
gqi.init(voxel);
deconvolution.init(voxel);
}
if (voxel.odf_decomposition)
{
gqi.init(voxel);
decomposition.init(voxel);
}
angle_variance = 8; //degress;
angle_variance *= M_PI/180;
angle_variance *= angle_variance;
angle_variance *= 2.0;
}