Alternative Trajectories

[johannesmayer]

The following trajectories are rather easy to simulate:

• variable density cartesian with golden cut angle increment (this is easy if one alread has a file, but can be simulated relatively easy by ordering the cartesian acquisitions in the correct way) -> check literature of Doneva Poisson Disk sampling or Küstner or Lustig et al. for formulas.

The following 'non-cartesian' trajectories need to be enabled:

• radial readout / stack-of-stars
• RPE
• real 3D encoding / arbitrary sampling trajectories For this couple the encoding to The Gadgetron software.

Select encoding based on the trajectory parameter set in the ISMRMRD Header passed to the simulation. Setting this will be enabled in the interface of the python/matlab code .

[johannesmayer]

In order to provide RPE support, one needs NFFT encoding: The interface is given in the toolbox <gadgetron/hoNFFT.h> Use is exemplified in CPUGriddingReconGadget:

hoNDArray<float_complext> *data;
hoNDArray<floatd2> *traj; // this has to be 2D since we do cartesian readouts
hoNDArray<float> *dcw; // this has to be 0x0 array because we forward project

hoNFFT_plan<float, 2> plan( from_std_vector<size_t, 2>(imageDims),
                               oversamplingFactor,
                                           kernelWidth);    

hoNDArray<float_complext> result; result.create(imageDimsOs[0], imageDimsOs[1]);

plan.preprocess(*traj);

// last argument has to be forward with Cartesian to non Carteisan layout.
plan.compute(*data, result, *dcw, hoNFFT_plan<float, 2>::NFFT_BACKWARDS_NC2C); 
return boost::make_shared<hoNDArray<float_complext>>(result); // or any other post processing of result!

[johannesmayer]

In vector_td_utilities the function

template<class T, unsigned int D> inline
vector_td<T,D> from_std_vector( std::vector<T> &_vector )

provides interface to std vector extraction of dimensions

[johannesmayer]

Check influence of parameter oversampling_factor in encoding call hoNFFT_plan<float, 2> nufft_operator( from_std_vector<size_t, 2>(slice_dims) , oversampling_factor, kernel_size);

This has to do with deapodization but I don't know how yet.

[johannesmayer]

Rewrite encoding to take 4d data since coilmaps take up an extra dimension. Dimensions need then be [N_ro, N_r, N_ang, N_coil]

[johannesmayer]

Provide automated trajectory computing based on input files. Slice and Phase encoding then become parameters of the trajectory. This would allow trajectory consistency with the rawdata files at all times.

[johannesmayer]

Interace is provided now. Now a radial readout should be enabled!

[johannesmayer]

Enable subtrajectory sampling, s.t. only points which necessary are actually interpolated