static Image::Header param (const Image::Header& header, size_t nvols) {
Image::Header ret (header);
ret.datatype() = DataType::Float32;
if (nvols) ret.dim(3) = nvols;
else ret.set_ndim (3);
return ret;
}
Dynamic::~Dynamic()
{
INFO ("Dynamic seeeding required " + str (total_samples) + " samples to draw " + str (total_seeds) + " seeds");
#ifdef DYNAMIC_SEED_DEBUGGING
const double final_mu = mu();
// Output seeding probabilites at end of execution
Image::Header H;
H.info() = info();
H.datatype() = DataType::Float32;
Image::Buffer<float> prob_mean_data ("seed_prob_mean.mif", H), prob_sum_data ("seed_prob_sum.mif", H);
Image::Buffer<float>::voxel_type prob_mean (prob_mean_data), prob_sum (prob_sum_data);
VoxelAccessor v (accessor);
Image::Loop loop;
for (loop.start (v, prob_mean, prob_sum); loop.ok(); loop.next (v, prob_mean, prob_sum)) {
if (v.value()) {
float sum = 0.0;
size_t count = 0;
for (Fixel_map<Fixel_TD_seed>::ConstIterator i = begin (v); i; ++i) {
sum += i().get_seed_prob (final_mu);
++count;
}
prob_mean.value() = sum / float(count);
prob_sum .value() = sum;
}
}
#endif
}
void run ()
{
Image::Buffer<float> dir_buf (argument[0]);
Image::Buffer<float>::voxel_type dir_vox (dir_buf);
Image::Header header (argument[0]);
header.dim(3) = header.dim(3)/3;
Image::Buffer<float> amp_buf (argument[1], header);
Image::Buffer<float>::voxel_type amp_vox (amp_buf);
Image::LoopInOrder loop (dir_vox, "converting directions to amplitudes...", 0, 3);
for (loop.start (dir_vox, amp_vox); loop.ok(); loop.next (dir_vox, amp_vox)) {
Math::Vector<float> dir (3);
dir_vox[3] = 0;
amp_vox[3] = 0;
while (dir_vox[3] < dir_vox.dim(3)) {
dir[0] = dir_vox.value(); ++dir_vox[3];
dir[1] = dir_vox.value(); ++dir_vox[3];
dir[2] = dir_vox.value(); ++dir_vox[3];
float amplitude = 0.0;
if (std::isfinite (dir[0]) && std::isfinite (dir[1]) && std::isfinite (dir[2]))
amplitude = Math::norm (dir);
amp_vox.value() = amplitude;
++amp_vox[3];
}
}
}
void print_comments (const Image::Header& header)
{
std::string buffer;
for (std::vector<std::string>::const_iterator i = header.comments().begin(); i != header.comments().end(); ++i)
buffer += *i + "\n";
std::cout << buffer;
}
void run () {
Image::BufferPreload<bool> input_data (argument[0]);
Image::BufferPreload<bool>::voxel_type input_voxel (input_data);
const size_t filter_index = argument[1];
Image::Filter::Base* filter = NULL;
switch (filter_index) {
case 0: filter = create_dilate_filter (input_voxel); break;
case 1: filter = create_erode_filter (input_voxel); break;
case 2: filter = create_lcc_filter (input_voxel); break;
case 3: filter = create_median_filter (input_voxel); break;
default: assert (0);
}
Image::Header header;
header.info() = filter->info();
Image::Stride::set_from_command_line (header);
Image::Buffer<bool> output_data (argument[2], header);
Image::Buffer<bool>::voxel_type output_voxel (output_data);
filter->set_message (std::string("applying ") + std::string(argument[1]) + " filter to image " + std::string(argument[0]) + "... ");
switch (filter_index) {
case 0: (*dynamic_cast<Image::Filter::Dilate*> (filter)) (input_voxel, output_voxel); break;
case 1: (*dynamic_cast<Image::Filter::Erode*> (filter)) (input_voxel, output_voxel); break;
case 2: (*dynamic_cast<Image::Filter::LargestConnectedComponent*> (filter)) (input_voxel, output_voxel); break;
case 3: (*dynamic_cast<Image::Filter::Median*> (filter)) (input_voxel, output_voxel); break;
}
delete filter;
}
void print_properties (const Image::Header& header)
{
std::string buffer;
for (std::map<std::string, std::string>::const_iterator i = header.begin(); i != header.end(); ++i)
buffer += i->first + ": " + i->second + "\n";
std::cout << buffer;
}
void print_vox (const Image::Header& header)
{
std::string buffer;
for (size_t i = 0; i < header.ndim(); ++i) {
if (i) buffer += " ";
buffer += str (header.vox (i));
}
std::cout << buffer << "\n";
}
void print_strides (const Image::Header& header)
{
std::string buffer;
std::vector<ssize_t> strides (Image::Stride::get (header));
Image::Stride::symbolise (strides);
for (size_t i = 0; i < header.ndim(); ++i) {
if (i) buffer += " ";
buffer += header.stride (i) ? str (strides[i]) : "?";
}
std::cout << buffer << "\n";
}
inline Point<> get_bounds (const Image::Header& header, Point<int> corner, int axis, const Point<>& ref, const Point<>& d1, const Point<>& d2) {
const Math::Matrix<float>& T (header.transform());
Point<> pos (T(0,3), T(1,3), T(2,3));
for (size_t n = 0; n < 3; ++n)
if (corner[n])
pos += header.dim(n) * header.vox(n) * Point<> (T(0,n), T(1,n), T(2,n));
Point<> dir (T(0,axis), T(1,axis), T(2,axis));
Point<> ret = get_bounds (pos, dir, ref, d1, d2);
ret[2] /= -header.vox (axis) * header.dim (axis);
return ret;
}
void set_strides (Image::Header& header)
{
Options opt = get_options ("stride");
if (opt.size()) {
std::vector<int> strides = opt[0][0];
if (strides.size() > header.ndim())
throw Exception ("too many axes supplied to -stride option");
for (size_t n = 0; n < strides.size(); ++n)
header.stride(n) = strides[n];
}
}
void run ()
{
conv_t conversion = NONE;
Options opt = get_options ("convert");
if (opt.size()) {
switch (int(opt[0][0])) {
case 0: conversion = OLD; break;
case 1: conversion = NEW; break;
case 2:
#ifndef USE_NON_ORTHONORMAL_SH_BASIS
conversion = NEW;
#else
conversion = OLD;
#endif
break;
case 3: conversion = FORCE_OLDTONEW; break;
case 4: conversion = FORCE_NEWTOOLD; break;
default: assert (0); break;
}
}
for (std::vector<ParsedArgument>::const_iterator i = argument.begin(); i != argument.end(); ++i) {
const std::string path = *i;
Image::Header H (path);
if (H.ndim() != 4) {
WARN ("Image \"" + H.name() + "\" is not 4D and therefore cannot be an SH image");
continue;
}
const size_t lmax = Math::SH::LforN (H.dim(3));
if (!lmax) {
WARN ("Image \"" + H.name() + "\" does not contain enough volumes to be an SH image");
continue;
}
if (Math::SH::NforL (lmax) != size_t(H.dim(3))) {
WARN ("Image \"" + H.name() + "\" does not contain a number of volumes appropriate for an SH image");
continue;
}
if (!H.datatype().is_floating_point()) {
WARN ("Image \"" + H.name() + "\" does not use a floating-point data type and therefore cannot be an SH image");
continue;
}
if (H.datatype().bytes() == 4)
check_and_update<float> (H, conversion);
else
check_and_update<double> (H, conversion);
}
};
void run ()
{
Image::Header header (argument[0]);
header.datatype() = DataType::Float32;
header.set_ndim (4);
header.dim(3) = 3;
Image::Stride::set (header, Image::Stride::contiguous_along_axis (3));
Image::Buffer<float> warp_buffer (argument[1], header);
Image::ThreadedLoop ("initialising warp image...", warp_buffer, 0, 3)
.run (write_coordinates (warp_buffer), warp_buffer.voxel());
}
void save_bvecs_bvals (const Image::Header& header, const std::string& path)
{
std::string bvecs_path, bvals_path;
if (path.size() >= 5 && path.substr (path.size() - 5, path.size()) == "bvecs") {
bvecs_path = path;
bvals_path = path.substr (0, path.size() - 5) + "bvals";
} else if (path.size() >= 5 && path.substr (path.size() - 5, path.size()) == "bvals") {
bvecs_path = path.substr (0, path.size() - 5) + "bvecs";
bvals_path = path;
} else {
bvecs_path = path + "bvecs";
bvals_path = path + "bvals";
}
const Math::Matrix<float>& grad (header.DW_scheme());
Math::Matrix<float> G (grad.rows(), 3);
// rotate vectors from scanner space to image space
Math::Matrix<float> D (header.transform());
Math::Permutation p (4);
int signum;
Math::LU::decomp (D, p, signum);
Math::Matrix<float> image2scanner (4,4);
Math::LU::inv (image2scanner, D, p);
Math::Matrix<float> rotation = image2scanner.sub (0,3,0,3);
Math::Matrix<float> grad_G = grad.sub (0, grad.rows(), 0, 3);
Math::mult (G, float(0.0), float(1.0), CblasNoTrans, grad_G, CblasTrans, rotation);
// deal with FSL requiring gradient directions to coincide with data strides
// also transpose matrices in preparation for file output
std::vector<size_t> order = Image::Stride::order (header, 0, 3);
Math::Matrix<float> bvecs (3, grad.rows());
Math::Matrix<float> bvals (1, grad.rows());
for (size_t n = 0; n < G.rows(); ++n) {
bvecs(0,n) = header.stride(order[0]) > 0 ? G(n,order[0]) : -G(n,order[0]);
bvecs(1,n) = header.stride(order[1]) > 0 ? G(n,order[1]) : -G(n,order[1]);
bvecs(2,n) = header.stride(order[2]) > 0 ? G(n,order[2]) : -G(n,order[2]);
bvals(0,n) = grad(n,3);
}
bvecs.save (bvecs_path);
bvals.save (bvals_path);
}
void oversample_header (Image::Header& header, const std::vector<float>& voxel_size)
{
INFO ("oversampling header...");
Math::Matrix<float> transform (header.transform());
for (size_t j = 0; j != 3; ++j) {
for (size_t i = 0; i < 3; ++i)
header.transform()(i,3) += 0.5 * (voxel_size[j] - header.vox(j)) * transform(i,j);
header.dim(j) = std::ceil(header.dim(j) * header.vox(j) / voxel_size[j]);
header.vox(j) = voxel_size[j];
}
}
void run ()
{
Image::Header input_header (argument[0]);
Image::Buffer<bool> input_data (input_header);
Image::Buffer<bool>::voxel_type input_voxel (input_data);
Image::Filter::ConnectedComponents connected_filter(input_voxel);
Image::Header header (input_data);
header.info() = connected_filter.info();
Image::Buffer<int> output_data (argument[1], header);
Image::Buffer<int>::voxel_type output_vox (output_data);
Options opt = get_options ("axes");
std::vector<int> axes;
if (opt.size()) {
axes = opt[0][0];
for (size_t d = 0; d < input_data.ndim(); d++)
connected_filter.set_ignore_dim (d, true);
for (size_t i = 0; i < axes.size(); i++) {
if (axes[i] >= static_cast<int> (input_header.ndim()) || axes[i] < 0)
throw Exception ("axis supplied to option -ignore is out of bounds");
connected_filter.set_ignore_dim (axes[i], false);
}
}
opt = get_options ("largest");
if (opt.size())
connected_filter.set_largest_only (true);
opt = get_options ("connectivity");
if (opt.size())
connected_filter.set_26_connectivity(true);
connected_filter.set_message ("computing connected components...");
connected_filter (input_voxel, output_vox);
}
Segmented_FOD_receiver (const Image::Header& header, const DWI::Directions::Set& directions) :
H (header),
H_fixel (header),
dirs (directions),
lmax (0)
{
// aPSF does not have data for lmax > 10
lmax = std::min (size_t(10), Math::SH::LforN (header.dim(3)));
H.set_ndim (3);
H.DW_scheme().clear();
H_fixel.set_ndim (3);
H_fixel.DW_scheme().clear();
H_fixel.datatype() = DataType::UInt64;
H_fixel.datatype().set_byte_order_native();
H_fixel[Image::Sparse::name_key] = str(typeid(FixelMetric).name());
H_fixel[Image::Sparse::size_key] = str(sizeof(FixelMetric));
}
void generate_header (Image::Header& header, const std::string& tck_file_path, const std::vector<float>& voxel_size)
{
Tractography::Properties properties;
Tractography::Reader<float> file (tck_file_path, properties);
Streamline<float> tck;
size_t track_counter = 0;
Point<float> min_values ( INFINITY, INFINITY, INFINITY);
Point<float> max_values (-INFINITY, -INFINITY, -INFINITY);
{
ProgressBar progress ("creating new template image...", 0);
while (file (tck) && track_counter++ < MAX_TRACKS_READ_FOR_HEADER) {
for (std::vector<Point<float> >::const_iterator i = tck.begin(); i != tck.end(); ++i) {
min_values[0] = std::min (min_values[0], (*i)[0]);
max_values[0] = std::max (max_values[0], (*i)[0]);
min_values[1] = std::min (min_values[1], (*i)[1]);
max_values[1] = std::max (max_values[1], (*i)[1]);
min_values[2] = std::min (min_values[2], (*i)[2]);
max_values[2] = std::max (max_values[2], (*i)[2]);
}
++progress;
}
}
min_values -= Point<float> (3.0*voxel_size[0], 3.0*voxel_size[1], 3.0*voxel_size[2]);
max_values += Point<float> (3.0*voxel_size[0], 3.0*voxel_size[1], 3.0*voxel_size[2]);
header.name() = "tckmap image header";
header.set_ndim (3);
for (size_t i = 0; i != 3; ++i) {
header.dim(i) = std::ceil((max_values[i] - min_values[i]) / voxel_size[i]);
header.vox(i) = voxel_size[i];
header.stride(i) = i+1;
//header.set_units (i, Image::Axis::millimeters);
}
//header.set_description (0, Image::Axis::left_to_right);
//header.set_description (1, Image::Axis::posterior_to_anterior);
//header.set_description (2, Image::Axis::inferior_to_superior);
Math::Matrix<float>& M (header.transform());
M.allocate (4,4);
M.identity();
M(0,3) = min_values[0];
M(1,3) = min_values[1];
M(2,3) = min_values[2];
file.close();
}
void run ()
{
Image::Buffer<node_t> nodes_data (argument[0]);
Image::Buffer<node_t>::voxel_type nodes (nodes_data);
Node_map node_map;
load_lookup_table (node_map);
Options opt = get_options ("config");
if (opt.size()) {
if (node_map.empty())
throw Exception ("Cannot properly interpret connectome config file if no lookup table is provided");
ConfigInvLookup config;
load_config (opt[0][0], config);
// Now that we know the configuration, can convert the lookup table to reflect the new indices
// If no corresponding entry exists in the config file, then the node doesn't get coloured
Node_map new_node_map;
for (Node_map::iterator i = node_map.begin(); i != node_map.end(); ++i) {
ConfigInvLookup::const_iterator existing = config.find (i->second.get_name());
if (existing != config.end())
new_node_map.insert (std::make_pair (existing->second, i->second));
}
if (new_node_map.empty())
throw Exception ("Config file and parcellation lookup table do not appear to belong to one another");
new_node_map.insert (std::make_pair (0, Node_info ("Unknown", Point<uint8_t> (0, 0, 0), 0)));
node_map = new_node_map;
}
if (node_map.empty()) {
INFO ("No lookup table provided; colouring nodes randomly");
node_t max_index = 0;
Image::LoopInOrder loop (nodes);
for (loop.start (nodes); loop.ok(); loop.next (nodes)) {
const node_t index = nodes.value();
if (index > max_index)
max_index = index;
}
node_map.insert (std::make_pair (0, Node_info ("None", 0, 0, 0, 0)));
Math::RNG rng;
for (node_t i = 1; i <= max_index; ++i) {
Point<uint8_t> colour;
do {
colour[0] = rng.uniform_int (255);
colour[1] = rng.uniform_int (255);
colour[2] = rng.uniform_int (255);
} while (int(colour[0]) + int(colour[1]) + int(colour[2]) < 100);
node_map.insert (std::make_pair (i, Node_info (str(i), colour)));
}
}
Image::Header H;
H.info() = nodes_data.info();
H.set_ndim (4);
H.dim(3) = 3;
H.datatype() = DataType::UInt8;
H.comments().push_back ("Coloured parcellation image generated by label2colour");
Image::Buffer<uint8_t> out_data (argument[1], H);
Image::Buffer<uint8_t>::voxel_type out (out_data);
Image::LoopInOrder loop (nodes, "Colourizing parcellated node image... ");
for (loop.start (nodes, out); loop.ok(); loop.next (nodes, out)) {
const node_t index = nodes.value();
Node_map::const_iterator i = node_map.find (index);
if (i == node_map.end()) {
out[3] = 0; out.value() = 0;
out[3] = 1; out.value() = 0;
out[3] = 2; out.value() = 0;
} else {
const Point<uint8_t>& colour (i->second.get_colour());
out[3] = 0; out.value() = colour[0];
out[3] = 1; out.value() = colour[1];
out[3] = 2; out.value() = colour[2];
}
}
}
void run ()
{
if (get_options ("norealign").size())
Image::Header::do_not_realign_transform = true;
const bool format = get_options("format") .size();
const bool ndim = get_options("ndim") .size();
const bool dimensions = get_options("dimensions") .size();
const bool vox = get_options("vox") .size();
const bool dt_long = get_options("datatype_long") .size();
const bool dt_short = get_options("datatype_short").size();
const bool stride = get_options("stride") .size();
const bool offset = get_options("offset") .size();
const bool multiplier = get_options("multiplier") .size();
const bool comments = get_options("comments") .size();
const bool properties = get_options("properties") .size();
const bool transform = get_options("transform") .size();
const bool dwgrad = get_options("dwgrad") .size();
Options opt = get_options ("export_grad_mrtrix");
const std::string dw_out_mrtrix = opt.size() ? opt[0][0] : std::string();
opt = get_options ("export_grad_fsl");
std::string dw_out_fsl_bvecs, dw_out_fsl_bvals;
if (opt.size()) {
dw_out_fsl_bvecs = str(opt[0][0]);
dw_out_fsl_bvals = str(opt[0][1]);
}
if ((dw_out_mrtrix.size() || dw_out_fsl_bvecs.size()) && (argument.size() > 1))
throw Exception ("Can only export gradient table information to file if a single input image is provided");
const bool print_full_header = (!(format || ndim || dimensions || vox || dt_long || dt_short || stride
|| offset || multiplier || comments || properties || transform || dwgrad)
&& dw_out_mrtrix.empty() && dw_out_fsl_bvecs.empty());
for (size_t i = 0; i < argument.size(); ++i) {
Image::Header header (argument[i]);
if (format) std::cout << header.format() << "\n";
if (ndim) std::cout << header.ndim() << "\n";
if (dimensions) print_dimensions (header);
if (vox) print_vox (header);
if (dt_long) std::cout << (header.datatype().description() ? header.datatype().description() : "invalid") << "\n";
if (dt_short) std::cout << (header.datatype().specifier() ? header.datatype().specifier() : "invalid") << "\n";
if (stride) print_strides (header);
if (offset) std::cout << header.intensity_offset() << "\n";
if (multiplier) std::cout << header.intensity_scale() << "\n";
if (comments) print_comments (header);
if (properties) print_properties (header);
if (transform) std::cout << header.transform();
if (dwgrad) std::cout << header.DW_scheme();
if (dw_out_mrtrix.size()) {
if (!header.DW_scheme().is_set())
throw Exception ("no gradient information found within image \"" + header.name() + "\"");
header.DW_scheme().save (dw_out_mrtrix);
}
if (dw_out_fsl_bvecs.size()) {
if (!header.DW_scheme().is_set())
throw Exception ("no gradient information found within image \"" + header.name() + "\"");
DWI::save_bvecs_bvals (header, dw_out_fsl_bvecs, dw_out_fsl_bvals);
}
if (print_full_header)
std::cout << header.description();
}
}
void run ()
{
Image::Buffer<value_type> SH_data (argument[0]);
Math::SH::check (SH_data);
Options opt = get_options ("mask");
std::unique_ptr<Image::Buffer<bool> > mask_data;
if (opt.size())
mask_data.reset (new Image::Buffer<bool> (opt[0][0]));
opt = get_options ("seeds");
Math::Matrix<value_type> dirs;
if (opt.size())
dirs.load (opt[0][0]);
else {
dirs.allocate (60,2);
dirs = Math::Matrix<value_type> (default_directions, 60, 2);
}
if (dirs.columns() != 2)
throw Exception ("expecting 2 columns for search directions matrix");
opt = get_options ("num");
int npeaks = opt.size() ? opt[0][0] : 3;
opt = get_options ("direction");
std::vector<Direction> true_peaks;
for (size_t n = 0; n < opt.size(); ++n) {
Direction p (Math::pi*to<float> (opt[n][0]) /180.0, Math::pi*float (opt[n][1]) /180.0);
true_peaks.push_back (p);
}
if (true_peaks.size())
npeaks = true_peaks.size();
opt = get_options ("threshold");
value_type threshold = -INFINITY;
if (opt.size())
threshold = opt[0][0];
Image::Header header (SH_data);
header.datatype() = DataType::Float32;
opt = get_options ("peaks");
std::unique_ptr<Image::Buffer<value_type> > ipeaks_data;
if (opt.size()) {
if (true_peaks.size())
throw Exception ("you can't specify both a peaks file and orientations to be estimated at the same time");
if (opt.size())
ipeaks_data.reset (new Image::Buffer<value_type> (opt[0][0]));
Image::check_dimensions (header, *ipeaks_data, 0, 3);
npeaks = ipeaks_data->dim (3) / 3;
}
header.dim(3) = 3 * npeaks;
Image::Buffer<value_type> peaks_data (argument[1], header);
DataLoader loader (SH_data, mask_data.get());
Processor processor (peaks_data, dirs, Math::SH::LforN (SH_data.dim (3)),
npeaks, true_peaks, threshold, ipeaks_data.get());
Thread::run_queue (loader, Item(), Thread::multi (processor));
}
void check_and_update (Image::Header& H, const conv_t conversion)
{
const size_t lmax = Math::SH::LforN (H.dim(3));
// Flag which volumes are m==0 and which are not
const ssize_t N = H.dim(3);
BitSet mzero_terms (N, false);
for (size_t l = 2; l <= lmax; l += 2)
mzero_terms[Math::SH::index (l, 0)] = true;
// Open in read-write mode if there's a chance of modification
typename Image::Buffer<value_type> buffer (H, (conversion != NONE));
typename Image::Buffer<value_type>::voxel_type v (buffer);
// Need to mask out voxels where the DC term is zero
Image::Info info_mask (H);
info_mask.set_ndim (3);
info_mask.datatype() = DataType::Bit;
Image::BufferScratch<bool> mask (info_mask);
Image::BufferScratch<bool>::voxel_type v_mask (mask);
size_t voxel_count = 0;
{
Image::LoopInOrder loop (v, "Masking image based on DC term...", 0, 3);
for (loop.start (v, v_mask); loop.ok(); loop.next (v, v_mask)) {
const value_type value = v.value();
if (value && std::isfinite (value)) {
v_mask.value() = true;
++voxel_count;
} else {
v_mask.value() = false;
}
}
}
// Get sums independently for each l
// Each order has a different power, and a different number of m!=0 volumes.
// Therefore, calculate the mean-square intensity for the m==0 and m!=0
// volumes independently, and report ratio for each harmonic order
Ptr<ProgressBar> progress;
if (App::log_level > 0 && App::log_level < 2)
progress = new ProgressBar ("Evaluating SH basis of image \"" + H.name() + "\"...", N-1);
std::vector<float> ratios;
for (size_t l = 2; l <= lmax; l += 2) {
double mzero_sum = 0.0, mnonzero_sum = 0.0;
Image::LoopInOrder loop (v, 0, 3);
for (v[3] = ssize_t (Math::SH::NforL(l-2)); v[3] != ssize_t (Math::SH::NforL(l)); ++v[3]) {
double sum = 0.0;
for (loop.start (v, v_mask); loop.ok(); loop.next (v, v_mask)) {
if (v_mask.value())
sum += Math::pow2 (value_type(v.value()));
}
if (mzero_terms[v[3]]) {
mzero_sum += sum;
DEBUG ("Volume " + str(v[3]) + ", m==0, sum " + str(sum));
} else {
mnonzero_sum += sum;
DEBUG ("Volume " + str(v[3]) + ", m!=0, sum " + str(sum));
}
if (progress)
++*progress;
}
const double mnonzero_MSoS = mnonzero_sum / (2.0 * l);
const float power_ratio = mnonzero_MSoS/mzero_sum;
ratios.push_back (power_ratio);
INFO ("SH order " + str(l) + ", ratio of m!=0 to m==0 power: " + str(power_ratio) +
", m==0 power: " + str (mzero_sum));
}
if (progress)
progress = NULL;
// First is ratio to be used for SH basis decision, second is gradient of regression
std::pair<float, float> regression = std::make_pair (0.0f, 0.0f);
size_t l_for_decision;
float power_ratio;
// The gradient will change depending on the current basis, so the threshold needs to also
// The gradient is as a function of l, not of even orders
float grad_threshold = 0.02;
switch (lmax) {
// Lmax == 2: only one order to use
case 2:
power_ratio = ratios.front();
l_for_decision = 2;
break;
// Lmax = 4: Use l=4 order to determine SH basis, can't check gradient since l=2 is untrustworthy
case 4:
power_ratio = ratios.back();
l_for_decision = 4;
break;
// Lmax = 6: Use l=4 order to determine SH basis, but checking the gradient is not reliable:
// artificially double the threshold so the power ratio at l=6 needs to be substantially
// different to l=4 to throw a warning
case 6:
regression = std::make_pair (ratios[1] - 2*(ratios[2]-ratios[1]), 0.5*(ratios[2]-ratios[1]));
power_ratio = ratios[1];
l_for_decision = 4;
grad_threshold *= 2.0;
break;
// Lmax >= 8: Do a linear regression from l=4 to l=lmax, project back to l=0
// (this is a more reliable quantification on poor data than l=4 alone)
default:
regression = get_regression (ratios);
power_ratio = regression.first;
l_for_decision = 0;
break;
}
// If the gradient is in fact positive (i.e. power ration increases for larger l), use the
// regression to pull the power ratio from l=lmax
if (regression.second > 0.0) {
l_for_decision = lmax;
power_ratio = regression.first + (lmax * regression.second);
}
DEBUG ("Power ratio for assessing SH basis is " + str(power_ratio) + " as " + (lmax < 8 ? "derived from" : "regressed to") + " l=" + str(l_for_decision));
// Threshold to make decision on what basis the data are currently stored in
value_type multiplier = 1.0;
if ((power_ratio > (5.0/3.0)) && (power_ratio < (7.0/3.0))) {
CONSOLE ("Image \"" + str(H.name()) + "\" appears to be in the old non-orthonormal basis");
switch (conversion) {
case NONE: break;
case OLD: break;
case NEW: multiplier = 1.0 / M_SQRT2; break;
case FORCE_OLDTONEW: multiplier = 1.0 / M_SQRT2; break;
case FORCE_NEWTOOLD: WARN ("Refusing to convert image \"" + H.name() + "\" from new to old basis, as data appear to already be in the old non-orthonormal basis"); return;
}
grad_threshold *= 2.0;
} else if ((power_ratio > (2.0/3.0)) && (power_ratio < (4.0/3.0))) {
CONSOLE ("Image \"" + str(H.name()) + "\" appears to be in the new orthonormal basis");
switch (conversion) {
case NONE: break;
case OLD: multiplier = M_SQRT2; break;
case NEW: break;
case FORCE_OLDTONEW: WARN ("Refusing to convert image \"" + H.name() + "\" from old to new basis, as data appear to already be in the new orthonormal basis"); return;
case FORCE_NEWTOOLD: multiplier = M_SQRT2; break;
}
} else {
multiplier = 0.0;
WARN ("Cannot make unambiguous decision on SH basis of image \"" + H.name()
+ "\" (power ratio " + (lmax < 8 ? "in" : "regressed to") + " " + str(l_for_decision) + " is " + str(power_ratio) + ")");
if (conversion == FORCE_OLDTONEW) {
WARN ("Forcing conversion of image \"" + H.name() + "\" from old to new SH basis on user request; however NO GUARANTEE IS PROVIDED on appropriateness of this conversion!");
multiplier = 1.0 / M_SQRT2;
} else if (conversion == FORCE_NEWTOOLD) {
WARN ("Forcing conversion of image \"" + H.name() + "\" from new to old SH basis on user request; however NO GUARANTEE IS PROVIDED on appropriateness of this conversion!");
multiplier = M_SQRT2;
}
}
// Decide whether the user needs to be warned about a poor diffusion encoding scheme
if (regression.second)
DEBUG ("Gradient of regression is " + str(regression.second) + "; threshold is " + str(grad_threshold));
if (Math::abs(regression.second) > grad_threshold) {
WARN ("Image \"" + H.name() + "\" may have been derived from poor directional encoding, or have some other underlying data problem");
WARN ("(m!=0 to m==0 power ratio changing by " + str(2.0*regression.second) + " per even order)");
}
// Adjust the image data in-place if necessary
if (multiplier && (multiplier != 1.0)) {
Image::LoopInOrder loop (v, 0, 3);
ProgressBar progress ("Modifying SH basis of image \"" + H.name() + "\"...", N-1);
for (v[3] = 1; v[3] != N; ++v[3]) {
if (!mzero_terms[v[3]]) {
for (loop.start (v); loop.ok(); loop.next (v))
v.value() *= multiplier;
}
++progress;
}
} else if (multiplier && (conversion != NONE)) {
INFO ("Image \"" + H.name() + "\" already in desired basis; nothing to do");
}
}
void run () {
Tractography::Properties properties;
Tractography::Reader<float> file (argument[0], properties);
const size_t num_tracks = properties["count"].empty() ? 0 : to<size_t> (properties["count"]);
float step_size = 0.0;
if (properties.find ("output_step_size") != properties.end())
step_size = to<float> (properties["output_step_size"]);
else
step_size = to<float> (properties["step_size"]);
std::vector<float> voxel_size;
Options opt = get_options("vox");
if (opt.size())
voxel_size = opt[0][0];
if (voxel_size.size() == 1)
voxel_size.assign (3, voxel_size.front());
else if (!voxel_size.empty() && voxel_size.size() != 3)
throw Exception ("voxel size must either be a single isotropic value, or a list of 3 comma-separated voxel dimensions");
if (!voxel_size.empty())
INFO ("creating image with voxel dimensions [ " + str(voxel_size[0]) + " " + str(voxel_size[1]) + " " + str(voxel_size[2]) + " ]");
Image::Header header;
opt = get_options ("template");
if (opt.size()) {
Image::Header template_header (opt[0][0]);
header = template_header;
header.comments().clear();
if (!voxel_size.empty())
oversample_header (header, voxel_size);
}
else {
if (voxel_size.empty())
throw Exception ("please specify either a template image or the desired voxel size");
generate_header (header, argument[0], voxel_size);
}
header.set_ndim (3);
opt = get_options ("contrast");
contrast_t contrast = opt.size() ? contrast_t(int(opt[0][0])) : TDI;
opt = get_options ("stat_vox");
vox_stat_t stat_vox = opt.size() ? vox_stat_t(int(opt[0][0])) : V_SUM;
opt = get_options ("stat_tck");
tck_stat_t stat_tck = opt.size() ? tck_stat_t(int(opt[0][0])) : T_MEAN;
float gaussian_fwhm_tck = 0.0, gaussian_denominator_tck = 0.0;
opt = get_options ("fwhm_tck");
if (opt.size()) {
if (stat_tck != GAUSSIAN) {
INFO ("Overriding per-track statistic to Gaussian as a full-width half-maximum has been provided.");
stat_tck = GAUSSIAN;
}
gaussian_fwhm_tck = opt[0][0];
const float gaussian_theta_tck = gaussian_fwhm_tck / (2.0 * sqrt (2.0 * log (2.0)));
gaussian_denominator_tck = 2.0 * gaussian_theta_tck * gaussian_theta_tck;
} else if (stat_tck == GAUSSIAN) {
throw Exception ("If using Gaussian per-streamline statistic, need to provide a full-width half-maximum for the Gaussian kernel using the -fwhm option");
}
opt = get_options ("colour");
const bool colour = opt.size();
opt = get_options ("map_zero");
const bool map_zero = opt.size();
if (colour) {
header.set_ndim (4);
header.dim(3) = 3;
//header.set_description (3, "directionally-encoded colour");
Image::Stride::set (header, Image::Stride::contiguous_along_axis (3, header));
}
// Deal with erroneous statistics & provide appropriate messages
switch (contrast) {
case TDI:
if (stat_vox != V_SUM && stat_vox != V_MEAN) {
INFO ("Cannot use voxel statistic other than 'sum' or 'mean' for TDI generation - ignoring");
stat_vox = V_SUM;
}
if (stat_tck != T_MEAN)
INFO ("Cannot use track statistic other than default for TDI generation - ignoring");
stat_tck = T_MEAN;
break;
case PRECISE_TDI:
if (stat_vox != V_SUM) {
INFO ("Cannot use voxel statistic other than 'sum' for precise TDI generation - ignoring");
stat_vox = V_SUM;
}
if (stat_tck != T_MEAN)
INFO ("Cannot use track statistic other than default for precise TDI generation - ignoring");
stat_tck = T_MEAN;
break;
case ENDPOINT:
if (stat_vox != V_SUM && stat_vox != V_MEAN) {
INFO ("Cannot use voxel statistic other than 'sum' or 'mean' for endpoint map generation - ignoring");
stat_vox = V_SUM;
}
if (stat_tck != T_MEAN)
INFO ("Cannot use track statistic other than default for endpoint map generation - ignoring");
stat_tck = T_MEAN;
break;
case LENGTH:
if (stat_tck != T_MEAN)
INFO ("Cannot use track statistic other than default for length-weighted TDI generation - ignoring");
stat_tck = T_MEAN;
break;
case INVLENGTH:
if (stat_tck != T_MEAN)
INFO ("Cannot use track statistic other than default for inverse-length-weighted TDI generation - ignoring");
stat_tck = T_MEAN;
break;
case SCALAR_MAP:
case SCALAR_MAP_COUNT:
break;
case FOD_AMP:
if (stat_tck == ENDS_MIN || stat_tck == ENDS_MEAN || stat_tck == ENDS_MAX || stat_tck == ENDS_PROD)
throw Exception ("Can't use endpoint-based track-wise statistics with FOD_AMP contrast");
break;
case CURVATURE:
break;
default:
throw Exception ("Undefined contrast mechanism");
}
opt = get_options ("datatype");
bool manual_datatype = false;
if (colour) {
header.datatype() = DataType::Float32;
manual_datatype = true;
}
if (opt.size()) {
if (colour) {
INFO ("Can't manually set datatype for directionally-encoded colour processing - overriding to Float32");
} else {
header.datatype() = DataType::parse (opt[0][0]);
manual_datatype = true;
}
}
if (get_options ("tck_weights_in").size() || (manual_datatype && header.datatype().is_integer())) {
if ((manual_datatype && header.datatype().is_integer()))
INFO ("Can't use an integer type if streamline weights are provided; overriding to Float32");
header.datatype() = DataType::Float32;
manual_datatype = true;
}
header.datatype().set_byte_order_native();
for (Tractography::Properties::iterator i = properties.begin(); i != properties.end(); ++i)
header.comments().push_back (i->first + ": " + i->second);
for (std::multimap<std::string,std::string>::const_iterator i = properties.roi.begin(); i != properties.roi.end(); ++i)
header.comments().push_back ("ROI: " + i->first + " " + i->second);
for (std::vector<std::string>::iterator i = properties.comments.begin(); i != properties.comments.end(); ++i)
header.comments().push_back ("comment: " + *i);
size_t upsample_ratio = 1;
opt = get_options ("upsample");
if (opt.size()) {
upsample_ratio = opt[0][0];
INFO ("track upsampling ratio manually set to " + str(upsample_ratio));
}
else if (step_size && std::isfinite (step_size)) {
// If accurately calculating the length through each voxel traversed, need a higher upsampling ratio
// (1/10th of the voxel size was found to give a good quantification of chordal length)
// For all other applications, making the upsampled step size about 1/3rd of a voxel seems sufficient
const float voxel_step_ratio = (contrast == PRECISE_TDI) ? 0.1 : 0.333;
upsample_ratio = Math::ceil<size_t> (step_size / (minvalue (header.vox(0), header.vox(1), header.vox(2)) * voxel_step_ratio));
INFO ("track upsampling ratio automatically set to " + str(upsample_ratio));
}
else {
if (contrast != PRECISE_TDI)
WARN ("track upsampling off; track step size information in header is absent or malformed");
}
const bool dump = get_options ("dump").size();
if (dump && Path::has_suffix (argument[1], "mih"))
throw Exception ("Option -dump only works when outputting to .mih image format");
std::string msg = str("Generating ") + (colour ? "colour " : "") + "image with ";
switch (contrast) {
case TDI: msg += "density"; break;
case PRECISE_TDI: msg += "density (precise)"; break;
case ENDPOINT: msg += "endpoint density"; break;
case LENGTH: msg += "length"; break;
case INVLENGTH: msg += "inverse length"; break;
case SCALAR_MAP: msg += "scalar map"; break;
case SCALAR_MAP_COUNT: msg += "scalar-map-thresholded tdi"; break;
case FOD_AMP: msg += "FOD amplitude"; break;
case CURVATURE: msg += "curvature"; break;
default: msg += "ERROR"; break;
}
msg += " contrast";
if (contrast == SCALAR_MAP || contrast == SCALAR_MAP_COUNT || contrast == FOD_AMP || contrast == CURVATURE)
msg += ", ";
else
msg += " and ";
switch (stat_vox) {
case V_SUM: msg += "summed"; break;
case V_MIN: msg += "minimum"; break;
case V_MEAN: msg += "mean"; break;
case V_MAX: msg += "maximum"; break;
default: msg += "ERROR"; break;
}
msg += " per-voxel statistic";
if (contrast == SCALAR_MAP || contrast == SCALAR_MAP_COUNT || contrast == FOD_AMP || contrast == CURVATURE) {
msg += " and ";
switch (stat_tck) {
case T_SUM: msg += "summed"; break;
case T_MIN: msg += "minimum"; break;
case T_MEAN: msg += "mean"; break;
case T_MAX: msg += "maximum"; break;
case T_MEDIAN: msg += "median"; break;
case T_MEAN_NONZERO: msg += "mean (nonzero)"; break;
case GAUSSIAN: msg += "gaussian (FWHM " + str (gaussian_fwhm_tck) + "mm)"; break;
case ENDS_MIN: msg += "endpoints (minimum)"; break;
case ENDS_MEAN: msg += "endpoints (mean)"; break;
case ENDS_MAX: msg += "endpoints (maximum)"; break;
case ENDS_PROD: msg += "endpoints (product)"; break;
default: msg += "ERROR"; break;
}
msg += " per-track statistic";
}
INFO (msg);
switch (contrast) {
case TDI: header.comments().push_back ("track density image"); break;
case PRECISE_TDI: header.comments().push_back ("track density image (precise calculation)"); break;
case ENDPOINT: header.comments().push_back ("track endpoint density image"); break;
case LENGTH: header.comments().push_back ("track density image (weighted by track length)"); break;
case INVLENGTH: header.comments().push_back ("track density image (weighted by inverse track length)"); break;
case SCALAR_MAP: header.comments().push_back ("track-weighted image (using scalar image)"); break;
case SCALAR_MAP_COUNT: header.comments().push_back ("track density image (using scalar image thresholding)"); break;
case FOD_AMP: header.comments().push_back ("track-weighted image (using FOD amplitude)"); break;
case CURVATURE: header.comments().push_back ("track-weighted image (using track curvature)"); break;
}
TrackLoader loader (file, num_tracks);
// Use a branching IF instead of a switch statement to permit scope
if (contrast == TDI || contrast == ENDPOINT || contrast == LENGTH || contrast == INVLENGTH || (contrast == CURVATURE && stat_tck != GAUSSIAN)) {
if (!manual_datatype) {
header.datatype() = (contrast == TDI || contrast == ENDPOINT) ? DataType::UInt32 : DataType::Float32;
header.datatype().set_byte_order_native();
}
if (colour) {
TrackMapperTWI <SetVoxelDEC> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck);
MapWriterColour<SetVoxelDEC> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDEC(), writer);
} else {
TrackMapperTWI <SetVoxel> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck);
Ptr< MapWriterBase<SetVoxel> > writer (make_writer<SetVoxel> (header, argument[1], dump, stat_vox));
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxel(), *writer);
}
} else if (contrast == PRECISE_TDI) {
if (!manual_datatype) {
header.datatype() = DataType::Float32;
header.datatype().set_byte_order_native();
}
if (colour) {
TrackMapperTWI <SetVoxelDir> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck);
MapWriterColour<SetVoxelDir> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDir(), writer);
} else {
TrackMapperTWI < SetVoxelDir> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck);
MapWriter <float, SetVoxelDir> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDir(), writer);
}
} else if (contrast == CURVATURE && stat_tck == GAUSSIAN) {
if (!manual_datatype) {
header.datatype() = DataType::Float32;
header.datatype().set_byte_order_native();
}
if (colour) {
TrackMapperTWI <SetVoxelDECFactor> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, gaussian_denominator_tck);
MapWriterColour<SetVoxelDECFactor> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDECFactor(), writer);
} else {
TrackMapperTWI <SetVoxelFactor> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, gaussian_denominator_tck);
Ptr< MapWriterBase<SetVoxelFactor> > writer (make_writer<SetVoxelFactor> (header, argument[1], dump, stat_vox));
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelFactor(), *writer);
}
} else if (contrast == SCALAR_MAP || contrast == SCALAR_MAP_COUNT || contrast == FOD_AMP) {
opt = get_options ("image");
if (!opt.size()) {
if (contrast == SCALAR_MAP || contrast == SCALAR_MAP_COUNT)
throw Exception ("If using 'scalar_map' or 'scalar_map_count' contrast, must provide the relevant scalar image using -image option");
else
throw Exception ("If using 'fod_amp' contrast, must provide the relevant spherical harmonic image using -image option");
}
Image::BufferPreload<float> input_image (opt[0][0]);
if ((contrast == SCALAR_MAP || contrast == SCALAR_MAP_COUNT)) {
if (!(input_image.ndim() == 3 || (input_image.ndim() == 4 && input_image.dim(3) == 1)))
throw Exception ("Use of 'scalar_map' contrast option requires a 3-dimensional image; your image is " + str(input_image.ndim()) + "D");
}
if (contrast == FOD_AMP && input_image.ndim() != 4)
throw Exception ("Use of 'fod_amp' contrast option requires a 4-dimensional image; your image is " + str(input_image.ndim()) + "D");
if (!manual_datatype && (input_image.datatype() != DataType::Bit))
header.datatype() = input_image.datatype();
if (colour) {
if (stat_tck == GAUSSIAN) {
TrackMapperTWIImage <SetVoxelDECFactor> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, gaussian_denominator_tck, input_image);
MapWriterColour <SetVoxelDECFactor> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDECFactor(), writer);
} else {
TrackMapperTWIImage <SetVoxelDEC> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, 0.0, input_image);
MapWriterColour <SetVoxelDEC> writer (header, argument[1], dump, stat_vox);
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelDEC(), writer);
}
} else {
if (stat_tck == GAUSSIAN) {
TrackMapperTWIImage <SetVoxelFactor> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, gaussian_denominator_tck, input_image);
Ptr< MapWriterBase <SetVoxelFactor> > writer (make_writer<SetVoxelFactor> (header, argument[1], dump, stat_vox));
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxelFactor(), *writer);
} else {
TrackMapperTWIImage <SetVoxel> mapper (header, upsample_ratio, map_zero, step_size, contrast, stat_tck, 0.0, input_image);
Ptr< MapWriterBase <SetVoxel> > writer (make_writer<SetVoxel> (header, argument[1], dump, stat_vox));
Thread::run_queue (loader, Tractography::Streamline<float>(), Thread::multi (mapper), SetVoxel(), *writer);
}
}
} else {
throw Exception ("Undefined contrast mechanism for output image");
}
}
void run () {
Point<> p[3];
for (size_t arg = 0; arg < 3; ++arg) {
std::vector<float> V = argument[arg];
if (V.size() != 3)
throw Exception ("coordinates must contain 3 elements");
p[arg][0] = V[0];
p[arg][1] = V[1];
p[arg][2] = V[2];
}
Image::Header header (argument[3]);
float vox = Math::pow (header.vox(0)*header.vox(1)*header.vox(2), 1.0f/3.0f);
Options opt = get_options ("vox");
if (opt.size())
vox = opt[0][0];
Point<> d1 = p[1] - p[0];
Point<> d2 = p[2] - p[0];
d1.normalise();
d2.normalise();
d2 = d2 - d1 * d1.dot(d2);
d2.normalise();
Point<> d3 = d1.cross(d2);
float min1 = std::numeric_limits<float>::infinity();
float min2 = std::numeric_limits<float>::infinity();
float max1 = -std::numeric_limits<float>::infinity();
float max2 = -std::numeric_limits<float>::infinity();
update_bounds (min1, min2, max1, max2, header, Point<int>(0,0,0), 0, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,0,0), 1, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,0,0), 2, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(1,0,0), 1, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(1,0,0), 2, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,1,0), 0, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,1,0), 2, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,0,1), 0, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,0,1), 1, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(1,1,0), 2, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(1,0,1), 1, p[0], d1, d2);
update_bounds (min1, min2, max1, max2, header, Point<int>(0,1,1), 0, p[0], d1, d2);
Point<> translation (p[0] + min1*d1 + min2*d2);
header.transform()(0,0) = d1[0];
header.transform()(1,0) = d1[1];
header.transform()(2,0) = d1[2];
header.transform()(0,1) = d2[0];
header.transform()(1,1) = d2[1];
header.transform()(2,1) = d2[2];
header.transform()(0,2) = d3[0];
header.transform()(1,2) = d3[1];
header.transform()(2,2) = d3[2];
header.transform()(0,3) = translation[0];
header.transform()(1,3) = translation[1];
header.transform()(2,3) = translation[2];
header.set_ndim (3);
header.dim (0) = (max1-min1) / vox;
header.dim (1) = (max2-min2) / vox;
header.dim (2) = 1;
header.vox(0) = header.vox(1) = header.vox(2) = vox;
header.datatype() = DataType::Bit;
header.DW_scheme().clear();
Image::Buffer<bool> roi_buffer (argument[4], header);
Image::Buffer<bool>::voxel_type roi (roi_buffer);
Image::LoopInOrder loop (roi);
for (loop.start (roi); loop.ok(); loop.next (roi))
roi.value() = true;
}
void run () {
Image::Header input_header (argument[0]);
const size_t filter_index = argument[1];
// Separate code path for FFT filter
if (!filter_index) {
// Gets preloaded by the FFT filter anyways, so just use a buffer
// FIXME Had to use cdouble throughout; seems to fail at compile time even trying to
// convert between cfloat and cdouble...
Image::Buffer<cdouble> input_data (input_header);
Image::Buffer<cdouble>::voxel_type input_voxel (input_data);
Image::Filter::FFT* filter = (dynamic_cast<Image::Filter::FFT*> (create_fft_filter (input_voxel)));
Image::Header header;
header.info() = filter->info();
set_strides (header);
filter->set_message (std::string("applying FFT filter to image " + std::string(argument[0]) + "..."));
if (get_options ("magnitude").size()) {
Image::BufferScratch<cdouble> temp_data (header, "complex FFT result");
Image::BufferScratch<cdouble>::voxel_type temp_voxel (temp_data);
(*filter) (input_voxel, temp_voxel);
header.datatype() = DataType::Float32;
Image::Buffer<float> output_data (argument[2], header);
Image::Buffer<float>::voxel_type output_voxel (output_data);
Image::LoopInOrder loop (output_voxel);
for (loop.start (temp_voxel, output_voxel); loop.ok(); loop.next (temp_voxel, output_voxel))
output_voxel.value() = std::abs (cdouble(temp_voxel.value()));
} else {
Image::Buffer<cdouble> output_data (argument[2], header);
Image::Buffer<cdouble>::voxel_type output_voxel (output_data);
(*filter) (input_voxel, output_voxel);
}
delete filter;
return;
}
Image::BufferPreload<float> input_data (input_header);
Image::BufferPreload<float>::voxel_type input_voxel (input_data);
Image::Filter::Base* filter = NULL;
switch (filter_index) {
case 0: assert (0); break;
case 1: filter = create_gradient_filter (input_voxel); break;
case 2: filter = create_median_filter (input_voxel); break;
case 3: filter = create_smooth_filter (input_voxel); break;
default: assert (0);
}
Image::Header header;
header.info() = filter->info();
set_strides (header);
Image::Buffer<float> output_data (argument[2], header);
Image::Buffer<float>::voxel_type output_voxel (output_data);
filter->set_message (std::string("applying ") + std::string(argument[1]) + " filter to image " + std::string(argument[0]) + "...");
switch (filter_index) {
case 0: assert (0); break;
case 1: (*dynamic_cast<Image::Filter::Gradient*> (filter)) (input_voxel, output_voxel); break;
case 2: (*dynamic_cast<Image::Filter::Median*> (filter)) (input_voxel, output_voxel); break;
case 3: (*dynamic_cast<Image::Filter::Smooth*> (filter)) (input_voxel, output_voxel); break;
}
delete filter;
}
Base::Base (const Image::Header& header) :
name (header.name()),
datatype (header.datatype()),
segsize (Image::voxel_count (header)),
is_new (false),
writable (false) { }
void run() {
Image::BufferPreload<float> data_in (argument[0], Image::Stride::contiguous_along_axis (3));
auto voxel_in = data_in.voxel();
Math::Matrix<value_type> grad (DWI::get_valid_DW_scheme<float> (data_in));
// Want to support non-shell-like data if it's just a straight extraction
// of all dwis or all bzeros i.e. don't initialise the Shells class
std::vector<size_t> volumes;
bool bzero = get_options ("bzero").size();
Options opt = get_options ("shell");
if (opt.size()) {
DWI::Shells shells (grad);
shells.select_shells (false, false);
for (size_t s = 0; s != shells.count(); ++s) {
DEBUG ("Including data from shell b=" + str(shells[s].get_mean()) + " +- " + str(shells[s].get_stdev()));
for (std::vector<size_t>::const_iterator v = shells[s].get_volumes().begin(); v != shells[s].get_volumes().end(); ++v)
volumes.push_back (*v);
}
// Remove DW information from header if b=0 is the only 'shell' selected
bzero = (shells.count() == 1 && shells[0].is_bzero());
} else {
const float bzero_threshold = File::Config::get_float ("BValueThreshold", 10.0);
for (size_t row = 0; row != grad.rows(); ++row) {
if ((bzero && (grad (row, 3) < bzero_threshold)) || (!bzero && (grad (row, 3) > bzero_threshold)))
volumes.push_back (row);
}
}
if (volumes.empty())
throw Exception ("No " + str(bzero ? "b=0" : "dwi") + " volumes present");
std::sort (volumes.begin(), volumes.end());
Image::Header header (data_in);
if (volumes.size() == 1)
header.set_ndim (3);
else
header.dim (3) = volumes.size();
Math::Matrix<value_type> new_grad (volumes.size(), grad.columns());
for (size_t i = 0; i < volumes.size(); i++)
new_grad.row (i) = grad.row (volumes[i]);
header.DW_scheme() = new_grad;
Image::Buffer<value_type> data_out (argument[1], header);
auto voxel_out = data_out.voxel();
Image::Loop outer ("extracting volumes...", 0, 3);
if (voxel_out.ndim() == 4) {
for (auto i = outer (voxel_out, voxel_in); i; ++i) {
for (size_t i = 0; i < volumes.size(); i++) {
voxel_in[3] = volumes[i];
voxel_out[3] = i;
voxel_out.value() = voxel_in.value();
}
}
} else {
const size_t volume = volumes[0];
for (auto i = outer (voxel_out, voxel_in); i; ++i) {
voxel_in[3] = volume;
voxel_out.value() = voxel_in.value();
}
}
}
