Comparison of iterative parametric and indirect deep learning-based reconstruction methods in highly undersampled DCE‐MR Imaging of the breast

These Matlab & Python codes are used as part of the work presented in:

Aditya Rastogi and Phaneendra K. Yalavarthy, “Comparison of iterative parametric and indirect deep learning-based reconstruction methods in highly undersampled DCE‐MR Imaging of the breast," Medical Physics, (2020), https://doi.org/10.1002/mp.14447

Due to the size of the code, testing data (Patient B) and trained model weights the code is uploaded on Google Drive. The link to the code is attached below:-

https://drive.google.com/drive/folders/12Byg-207xGgg6Sq13eYl2l-VGejXD7SV?usp=sharing

Please mail me at adityar[at]iisc[dot]ac[dot]in if you encounter any problem in Downloading, Executing or understanding the code.

The code has 4 folders:-

1. Test_Direct :- Constructs the Ktrans map using iterative direct reconstruction techniques.This generates the following
   • Fully sampled Ktrans (ground truth)
   • Ktrans with zero padding (US)
   • Ktrans with no regularization (L2), TV+L1 regularisation, L1 regularization only and TV regularization only.
   • This code is based on work done by Yi. Guo¹ in this paper "Direct Estimation of Tracer-Kinetic Parameter Maps from Highly Undersampled Brain DCE-MRI" and uses some codes and libraries from his program available at "https://github.com/usc-mrel/DCE_direct_recon".
   • This folder has 4 main files/folder:-
     1. main.m :- This file executes the code and estimates Ktrans map for undersampling rate of 20X, 50X and 100X.
     2. lam_mat.mat :- This .mat file contains the regularization parameter values of all methods for all undersampling rates.
     3. Dataset :- This folder contains the dataset of patient B.
     4. Vol :- This folder contains the recontructed Ktrans map of patient B for all undersampling rates (R).
2. Test_NN :- This folder contains the DL based models for indirect reconstruction of Ktrans maps. This code contains three folders:-
   • ISTA-Net_plus :- This folder contains the weights and testing file of ISTA-Net^+[2] as mentioned in paper :- "ISTA-Net: Interpretable Optimization-Inspired Deep Network for Image Compressive Sensing". This folder contains testing model and files for 20X, 50X and 100X undersampling. The test data of patient B and undersampling mask are present in folder of MODL. This code is used to estimate high resolution anatomical images from undersampled K-t space Data
     • MODL:- This folder contains the trained models for 20X,50X and 100X undersampling. This code is used to estimate high resolution anatomical images from undersampled K-t space Data. This method is give by Hemant Kumar Aggarwal in his paper "MoDL: Model Based Deep Learning Architecture for Inverse Problems" and the original code of the paper is available at https://github.com/hkaggarwal/modl .
     1. This folder also contains a folder name test_datasets which has the testing dataset of Patient B and the 20X, 50X and 100X undersampling masks.
   • TK_modelling:- This folder has 4 main components:-
     1. recon_NN :- This folder contains the .h5py file that is reconstructed from MODL and ISTA-Net⁺ .
     2. vol :- This folder contains the estimated Ktrans maps using the reconstructed anatomical images of MODL and ISTA-Net⁺ .
     3. Vol_ISTA_NN_Kt_Vp_SEN_AD_3d.m :- This file estimates Ktrans map from reconstructed anatomical images (via ISTA-Net⁺ ).
        1. Vol_MODL_NN_Kt_Vp_SEN_AD_3d.m :- This file estimates Ktrans map from reconstructed anatomical images (via MODL ).
3. Generate Results:- This folder contains code and data to compare the results of direct and indirect estimation techniques. This folder has 3 main files/folders:-
   • compare.m :- This file compares the reconstructed Ktrans map from direct reconstruction techniques and indirect reconstruction techniques using 4 metrices (PSNR, nRMSE, SSIM and Xydeas metric).
     • Vol_NN :- This folder contains the Ktrans maps reconstructed using indirect DL based techniques for all R.
     • Vol :- This folder contains the recontructed Ktrans map of patient B for all undersampling rates (R).
4. Make_plots :- This folder plots the barchart of performance of direct and indirect estimation techniques for Patient B. It has a folder:-
   • datasets:- This folder contains a .mat file which consists of performance results of the US, L2, TV+L1 in terms of PSNR, nRMSE, SSIM and Xydeas metric.
   • barplot.py :- Is the python execution file.