online STCR

Online reconstruction of real-time MR images using spatiotemporally constrained reconstruction (online STCR)

Data

Please download the fetal real-time raw k-space data and add to the main folder.

Usage

Running Inputs:

1. Download the Parallel Computing toolbox for MATLAB if not already available.

2. Clone the repository with the submodules mirt and usc_dynamic_reconstruction by typing into your terminal

   git clone --recurse-submodules https://github.com/usc-mrel/onlineSTCR

   This will automatically add folders mirt and usc_dynamic_reconstruction in your main path.

3. Make sure to put the downloaded fetal real-time raw-kspace data in your main path.

4. Set the relevant parameters and flags inside

   usc_dynamic_reconstruction/recon_online_stcr.m
   

   including

   Narms_full: Number of TRs binned for full sampling to estimate coil sensitivity

   Narms_recon: Desired temporal resolution (distance between frames)

   Narms_window: Numbers of TRs binned in 1 frame

   Narms_initial: Number of TRs binned for initial frame

   lambdaTFD: Temporal regularization parameter

   lambdatTV: Spatial regularization parameter

   mu: Penalty parameter

   step_size_x: Step size of shrinkage operator

   Nmaxiter: Number of iterations per frame. Tune this so that the time to reconstruct 1 frame is less than acquisition time.

   print_cost: Choose wheter to print the components of the cost function for each frame (will increase runtime)

   accelerate_flag: Choose whether to apply Nesterov acceleration in FISTA

   toeplitz_flag: Choose whether to use Toeplitz trick to reduce gridding and inverse-gridding time.

5. Then run

   main.m
   

Outputs:

Retrospective online STCR reconstruction returns real-time MR image in x, of shape [Nx x Ny x Nframes].

If print_cost = true, the cost of the online STCR optimization problem of each frame will be printed after every iteration.

Notes:

The code is tested with MATLAB 2021a on a server equipped with 4x NVIDIA A100 GPU (40GB memory for each).