Course

Eleven lessons that take you from a k-space tensor to an image, then through the reconstructions BART is known for, and finally to writing and training your own. Read them in order: each one starts from what the previous one built.

Lesson	Contents
1. Your data as tensors	Tensors in BART’s layout; BART commands on tensors
2. From k-space to an image	Undersampled Cartesian reconstruction, step by step
3. Coil sensitivities	Coil sensitivity estimation and its effect on the reconstruction
4. Joint reconstruction of image and sensitivity maps	Joint estimation of image and coil sensitivities from a small calibration region
5. Choosing a regularizer	Tikhonov, wavelet and total-variation regularization; regularization weight
6. Custom encoding and solver	Encoding operator and solver, identical to pics
7. Basics of non-Cartesian imaging	Trajectories, NUFFT, density compensation, point spread function
8. Radial SENSE	Golden-angle radial SENSE
9. Nonlinear inversion for T2 mapping	Model-based \(T_2\) mapping from raw data
10. Plug-and-play reconstruction	Plug-and-play with a pretrained denoiser
11. Training an unrolled network	Unrolled network trained through ADMM

By the end you can reconstruct Cartesian and non-Cartesian data with BART’s methods, write an encoding BART has no command for, and put a network inside the reconstruction.

Lessons 10 and 11 also need:

pip install lightning torchio monai deepinv

Lesson 10 downloads the DRUNet weights that deepinv distributes.