The Non-Uniform Fast Fourier Transform (NUFFT) appears a lot in MRI when designing non-Cartesian acquisition trajectories.
In order to design differentiable physics-based models, it is important to have a differentiable efficient implementation of the NUFFT.
This project is an offspring of the excellent work done by Matthew Muckley in PyTorch; torchkbnufft.
I was also happy to collaborate with Chaithya G R on this project, in particular receiving his help on density compensation and gradients with respect to the trajectory.
- fastMRI 2020 Brain MRI reconstruction Challenge
- Learning the sampling density in 2D SPARKLING MRI acquisition for optimized image reconstruction
- Benchmarking proximal methods acceleration enhancements for CS-acquired MR image analysis reconstruction
- Benchmarking MRI reconstruction neural networks on large public datasets