Speed issues: GRAPPA recon for host scan

[dgallichan]

It seems a shame to have to recreate the whole reconstruction pipeline that the scanner is doing anyway, but unfortunately there doesn't seem to be an easy way around this...

Currently GRAPPA is performed on each 'slice' after a 1D FFT in the readout direction - but the GRAPPA kernel is calculated using neighbouring slices as well to try to introduce some smoothness into the GRAPPA weights along the readout direction. Currently I speed this up by simply using a parfor loop over the slices - and this can reconstruct the 600 um resolution MP2RAGE data in about 5 minutes over 12 CPUs. However, the equivalent ICE reconstruction only takes a few seconds...

Perhaps the way forward here would be if an efficient 3D-GRAPPA code for Gadgetron becomes available then it would be worth putting the raw data into ISMRMRD format at the start of the retroMoCoBox pipeline.