search

MagneticResonanceImaging / MRIReco.jl

Julia Package for MRI Reconstruction
https://magneticresonanceimaging.github.io/MRIReco.jl/latest/
Other
85 stars 22 forks source link

add test for Bruker multichannel acquisition #57

Closed aTrotier closed 2 years ago

aTrotier commented 2 years ago

The test is working but the RMSE for 3D is high (> 10%) and for 2D (<2%)

If I check the plot I get for 2D : plot_83

And for 3D : plot_82

Maybe the scaling is part of the problem but the most important error is at the boundaries. I checked and I don't have an offset in the acquisition

##$PVM_SPackArrPhase1Offset=( 1 )
0
##$PVM_SPackArrPhase2Offset=( 1 )
0
##$PVM_SPackArrSliceOffset=( 1 )
0

Maybe an offset created by the fft in the 3rd direction ?

Any thoughts ?

By the way, the offset in phase/partition is not handle in the reconstruction I think ? I yes, should I open an issue as a reminder ? (I might have some codes for that under matlab)

tknopp commented 2 years ago

Its quite cool that this works out of the box.

I would say for now the accuracy issue is not so dramatic. The images look basically the same. Finding the root cause can be tricky since we do not know exactly what Bruker does. One thing one could do is throw in a third framework like BART and look if two reconstructions are closer together in the deviation. But that should not be included in the tests.

By the way, the offset in phase/partition is not handle in the reconstruction I think ? I yes, should I open an issue as a reminder ? (I might have some codes for that under matlab)

Yes please. And if you have such data it would be great to get this fixed. Probably it would be good to replace the testdata with something that has offset afterwards.

aTrotier commented 2 years ago

I would say for now the accuracy issue is not so dramatic. The images look basically the same. Finding the root cause can be tricky since we do not know exactly what Bruker does. One thing one could do is throw in a third framework like BART and look if two reconstructions are closer together in the deviation. But that should not be included in the tests.

Agreed I guess if we broke something later it will be far over 15% error :)

I also had some trouble with the BART reconstruction that creates an offset in comparison to the standard Siemens reconstruction.

tknopp commented 2 years ago

Merge this now, this is great Aurelien! Thank you for the contribution.