sensivity Operator input

[aTrotier]

Hi,
I am still playing around with the operator in order to well understand them before :)

I am able to reconstruct an coil-combined image from a multicoil non-cartesian acquisition :

nufftOp = MRIReco.NFFTOp((N, N), traj_MR)
im = Matrix(nufftOp'*(dcf.*raw_test))
sens = estimateCoilSensitivities(reshape(im,(300,300,1,1,12)))
#MIRTjim.jim(dropdims(abs.(sens),dims=(3,4)), title="Sensitivity")

sensOp = MRIReco.SensitivityOp(reshape(sens,(300*300,12)))

im_sens = sensOp' * vec(Matrix(nufftOp'*(dcf.*raw_test)))
#im_sens = sensOp' * vec(im)
p = Plots.heatmap(abs.(reshape(im_sens,(300,300))),color = :greys,aspect_ratio = 1)

The input are :

julia> nufftOp'*(dcf.*raw_test)
Linear operator
  nrow: 90000
  ncol: 12
  eltype: ComplexF64
  symmetric: false
  hermitian: false
  nprod:   0
  ntprod:  0
  nctprod: 0

julia> sensOp'
Adjoint of the following LinearOperator:
Linear operator
  nrow: 1080000
  ncol: 90000
  eltype: ComplexF64
  symmetric: false
  hermitian: false
  nprod:   12
  ntprod:  0
  nctprod: 3

My only concerns is that vec(matrix())operation that is required. Is there a way to make that more straightforward ?

I suppose it is a limitation from the LinearOperators package and the only to get rid of it will be to move to something else like :

• https://github.com/hakkelt/FunctionOperators.jl
• LinearMapsAA from @JeffFessler

[JeffFessler]

I think you are correct.

[tknopp]

Yes, LinearOperators right now only seem to support multiplications with vectors. It is really a downside that we have so many packages in that domain since it hinders us from mixing operator implementations.

[JeffFessler]

I'm adding some features to facilitate some degree of inter-operability:
https://github.com/JeffFessler/LinearMapsAA.jl#inter-operability

[aTrotier]

For sure, an harmonization between the MR toolbox can be great in order to mix the operator.