search

MagneticResonanceImaging / MRIReco.jl

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

Reco Bruker 3DUTE initial implementation #79

Closed aTrotier closed 2 years ago

aTrotier commented 2 years ago

This PR is not suppose to be merged for now

Goal

It is an initial implementation of a 3DUTE reconstruction from the standard 3D UTE Bruker acquisition.

Test

You can find a testing script here with the associated datasets (a 3DUTE with a matrix of 96x96x96 and 2 NR) https://drive.google.com/drive/folders/1roh0tRhAQTsqe27iKaHHhuGGi9n4MA65?usp=sharing

Results of test.jl :

RMSE = 0.30 test

Remarks

Discussions

1- Reconstruction only return 1 repetition (the for loop is not available for NR) 2- Only one trajectory is stored during conversion from raw to acq, should we keep one for each NR,contrast,slices ?

Let me know what are your thoughts about that first try.

codecov-commenter commented 2 years ago

Codecov Report

Merging #79 (eb6e54f) into master (efb3ac0) will increase coverage by 0.20%. The diff coverage is 0.00%.

:exclamation: Current head eb6e54f differs from pull request most recent head dadc68d. Consider uploading reports for the commit dadc68d to get more accurate results

@@            Coverage Diff             @@
##           master      #79      +/-   ##
==========================================
+ Coverage   63.87%   64.07%   +0.20%     
==========================================
  Files          37       37              
  Lines        1550     1545       -5     
==========================================
  Hits          990      990              
+ Misses        560      555       -5
Impacted Files Coverage Δ
src/Operators/Composition.jl 80.00% <ø> (ø)
src/Operators/EncodingOp.jl 50.84% <0.00%> (ø)
src/Operators/ExplicitOp.jl 92.30% <ø> (ø)
src/Operators/FieldmapNFFTOp.jl 87.85% <ø> (ø)
src/Operators/NFFTOp.jl 79.24% <ø> (ø)
src/Sampling/Sampling.jl 81.08% <ø> (+9.65%) :arrow_up:

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update efb3ac0...dadc68d. Read the comment docs.

aTrotier commented 2 years ago

Adding a circshift along the slice direction i get a rmse under 10 percent :

b = BrukerFile("data/UTE3D_NR2/")

raw = RawAcquisitionData_3DUTE(b)
acq = AcquisitionData(raw);  # TODO vérification des modifications qui ont étaient effectuée

params = Dict{Symbol, Any}()
params[:reco] = "direct"
if (acq.encodingSize[3]>1)
    params[:reconSize] = (acq.encodingSize[1],acq.encodingSize[2],acq.encodingSize[3]);
else
    params[:reconSize] = (acq.encodingSize[1],acq.encodingSize[2]);
end

Ireco = reconstruction(acq, params);
#@test size(Ireco) == (raw.params["encodedSize"][1], raw.params["encodedSize"][2], raw.params["encodedSize"][3], 1, raw.params["receiverChannels"])

Isos = sqrt.(sum(abs.(Ireco).^2,dims=5));
Isos = Isos ./ maximum(Isos);

I2dseq = recoData(b)
I2dseq = I2dseq ./ maximum(I2dseq);

# reorient
I2dseq = permutedims(I2dseq,(2,1,3,4))
I2dseq = circshift(I2dseq,(0,0,1,0))
#@test norm(vec(I2dseq)-vec(Isos))/norm(vec(I2dseq)) < listNormValues[i]

sl = 60
p1 =     Plots.heatmap(abs.(Isos[:,:,sl,1,1]),title = "MRIReco",color = :greys,aspect_ratio = 1, axis=nothing, lims=(1,95))
p2 =     Plots.heatmap(abs.(I2dseq[:,:,sl,1]),title = "Bruker",color = :greys,aspect_ratio = 1 , axis=nothing, lims=(1,95))

diff = abs.(abs.(I2dseq[:,:,sl,1]) - abs.(Isos[:,:,sl,1,1] )) ./ abs.(I2dseq[:,:,20,1])
p3 =     Plots.heatmap(diff,title = "Diff",color = :greys,aspect_ratio = 1 , axis=nothing, lims=(1,95))
Plots.plot(p1,p2,p3,layouts=(1,3))

MRIReco.norm(vec(I2dseq)-vec(Isos))/MRIReco.norm(vec(I2dseq))
tknopp commented 2 years ago

Great that we have support for more sequences. What I don't get is, why we need trajectoryCustom, rawdataCustom and RawAcquisitionData_3DUTE. Its fine that we have custom code but the dispatch should be done within the functions RawAcquisitionData, trajectory and rawdata.

aTrotier commented 2 years ago

It was just a first try to clearly separate the functions in order to discuss the implementation ?

If you are ok with it I can regroup that under RawAcquisitionData, trajectory and rawdata.

aTrotier commented 2 years ago

Can you upload the dataset in order to create a test ?

tknopp commented 2 years ago

Can you upload the dataset in order to create a test ?

Yes, will do later this week.

aTrotier commented 2 years ago

Thanks

tknopp commented 2 years ago

The data is now included, see https://github.com/MagneticResonanceImaging/MRIReco.jl/commit/eb6e54f67732c1247b3f1be6d64db0a602bdcd8b It is in the folder BrukerFile/3D_UTE_NR2/ To test this locally, you need to dev MRIFiles/MRIBase as well.

aTrotier commented 2 years ago

Thanks, with dev MRIFiles/MRIBase my test is working. Should I change the version of MRIFiles/MRIBase in order to make the CI tests works ?

tknopp commented 2 years ago

No, I will do this. Are you already using the artifact?

aTrotier commented 2 years ago

yes. I need to dev MRIFiles and MRIBase. The CI is not using the version in that PR. Am I missing something to make it work without making a PR of first MRIBase/MRIFiles ?

tknopp commented 2 years ago

Am I missing something to make it work without making a PR of first MRIBase/MRIFiles ?

no. There is no chance that you can make CI happy in this setting. I will try out locally and then do the releases.

tknopp commented 2 years ago

I made the releases but CI is only partially happy. It seems that there is probably just not enough memory for the UTE dataset on the windows CI: https://github.com/MagneticResonanceImaging/MRIReco.jl/runs/6240907107?check_suite_focus=true

aTrotier commented 2 years ago

Ok I will acquire a smaller dataset after ismrm

Le sam. 30 avr. 2022 à 18:06, Tobias Knopp @.***> a écrit :

I made the releases but CI is only partially happy. It seems that there is probably just not enough memory for the UTE dataset on the windows CI: https://github.com/MagneticResonanceImaging/MRIReco.jl/runs/6240907107?check_suite_focus=true

— Reply to this email directly, view it on GitHub https://github.com/MagneticResonanceImaging/MRIReco.jl/pull/79#issuecomment-1114012125, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC5P7O2AO56DHKAFCOG4STDVHVK67ANCNFSM5TK5KXVA . You are receiving this because you authored the thread.Message ID: @.***>