Roadmap discussion

[aTrotier]

Hello !

Since our last online meeting, we have implemented :

• A literate example for advanced reconstruction pipeline (binning data, extract kspace from acq and recreate acq struct from that) : works offline but does not show the image in the documentation... @aTrotier
• noise decorrelation @Laura2305
• Add different num of profiles along contrasts @aTrotier
• add pogm/optista solver to RegularizedLeastSquares.jl @andrewwmao @JeffFessler
• Geometric Coil Compression @aTrotier
• Dedicated package for MRIFieldmaps : @JeffFessler @alexjaffray
  and of course a lot of bug fixes 👍

Do you think we should work on a roadmap to focus our efforts ?
At least to know what we are working on or what we would like to see in the package.

Roadmap suggestion :

• Fixing the accuracy / convergence issues for CS Fista algorithm : accuracy problem #114
• implement and benchmark wavelet transformed for non-square / not 2^n matrix Maximul level transform for Wavelet operator in non 2^n matrix tknopp/SparsityOperators.jl#12
• Model-based reconstruction (moba in BART, PyQMRI)
• unified EPG package. for now we have multiple implementations in MRIReco and DECAES (other language implementation sycomore / EPGx) @jondeuce / @cncastillo / @JakobAsslaender

[cncastillo]

I just wanted to say that I will soon implement a GPU-accelerated EPG simulator (using KomaMRICore). I will start trying to match sycomore in terms of functionality, so without the exchange and magnetization transfer (EPG-X) in the first iteration. I am interested in this to generate dictionaries for MRF reconstruction. Can you point me to the implementation in MRIReco?

[aTrotier]

Hi, I have a simple epg implementation that mimic the sycomore implementation (for now only the regular case) and can be used with automatic differentiation : EPGsim.jl In contrary to the implementation in MRI simulation.jl, you don't need to define the whole epgstates array. I did not benchmark which implementation is the fastest one. I guess it will depends of the simulation. Le sam. 22 juil. 2023 à 00:38, Carlos Castillo Passi < ***@***.***> a écrit :

…

I just wanted to say that I will soon implement a GPU-accelerated EPG simulator (using KomaMRICore). I will start trying to match sycomore <https://sycomore.readthedocs.io/en/latest/index.html> in terms of functionality, so without the exchange and magnetization transfer (EPG-X) in the first iteration. I am interested in this to generate dictionaries for MRF reconstruction. Can you point me to the implementation in MRIReco? — Reply to this email directly, view it on GitHub <#149 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC5P7O7NUBNA3V5MIWWDFJLXRMAIDANCNFSM6AAAAAAZGD2H5Y> . You are receiving this because you authored the thread.Message ID: <MagneticResonanceImaging/MRIReco.jl/repo-discussions/149/comments/6513716 @github.com>

[tknopp]

Hey @cncastillo and @aTrotier,

it would be great if you could unify the existing code/packages into MRISimulation: https://github.com/MagneticResonanceImaging/MRIReco.jl/tree/master/MRISimulation

If you prefer, the package could also be developed outside of MRIReco.jl. Would just be great to have it in the MagneticResonanceImaging organization.

[aTrotier]

@cncastillo if you want I can move my project into the organization, I don't know if you can use that for a gpu implementatio.

The only requirement on my part is to be able to use AD for sequence optimization with Cramer-Rao Lower Bound. Maybe @JakobAsslaender also have some suggestions about that.
And I have another package that used it for T2 mapping of a MultiEcho Spin Echo sequence that use it. I am style trying to figure out how to reduce the time to fit an epg curve : https://github.com/aTrotier/QuantitativeMRI.jl

By the way, the maintainer of sycomore is one of my collaborator. We can ask him about the epg implementation details

@tknopp What are your thoughts about having a dedicated epg package and that MRISimulation only used them to create the signal for specific sequence ?

[aTrotier]

EPGsim.jl is now stored in the MagneticResonanceImaging organization

By the way @StevenWhitaker also has a package : https://github.com/StevenWhitaker/BlochSim.jl that also can be used for MR simulation

[tknopp]

I am not so deep into MRI simulation but I wonder if there would be the possibility to merge some of these packages into a single one.

[aTrotier]

I don't know if we should bring all of them into one package because they are not using the same mathematical/physics background.
But I suppose we can try to have at least 2 main packages for EPG and Bloch Simulation for the community that at implement common functionality of the following packages

Packages, I am aware of, that use MR simulation :
Extended Phase Graph implementation :

• https://github.com/jondeuce/DECAES.jl
• https://github.com/MagneticResonanceImaging/EPGsim.jl
• MRSimulation.jl

Bloch simulation :

• https://github.com/StevenWhitaker/BlochSim.jl
• https://github.com/JakobAsslaender/MRIgeneralizedBloch.jl
• https://github.com/cncastillo/KomaMRI.jl
• https://github.com/SpinDoctorMRI/SpinDoctor.jl

[aTrotier]

Another EPG implementation from @felixhorger : https://github.com/felixhorger/MRIEPG.jl