SimulatingDCE

Official repository of "Simulating Dynamic Tumor Contrast Enhancement in Breast MRI using Conditional Generative Adversarial Networks"

Getting Started

Synthesis Code

Multi-Sequence Synthesis

• Code to extract 2D pngs from 3D NiFti files for the multi-sequence train-val-test split.
• Code to concat multiple DCE-MRI sequences into channels of an image, which is later used as input into the synthesis model.
• Script to run a training of the version of the synthesis model that jointly generates images for corresponding DCE-MRI sequences 1, 2 and 3.
• Script to run a test (inference) of the trained synthesis model that will then jointly generates images for corresponding DCE-MRI sequences 1, 2 and 3 for the test set.
• Script Compute per DCE-MRI sequence (e.g. 1, 2 and 3) the imagenet and radimagenet-based Frèchet Inception Distances (FIDs) for the GAN generated data.
• Script Compute per DCE-MRI sequence (e.g. 1, 2 and 3) the imagenet and radimagenet-based Frèchet Inception Distances (FIDs) for the GAN generated data.
• Notebook to compute and visualize contrast enhancement kinetic patterns for corresponding multi-sequence DCE-MRI data.

Single-Sequence Synthesis

• Code to extract 2D pngs from 3D NiFti files for the single-sequence train-val-test split.
• Script to run a training of the image synthesis model.
• Script to run a test of the image synthesis model.

You may find some examples of synthetic nifti files in synthesis/examples.

General

• Code to transform Duke DICOM files to NiFti files.
• Code to create 3D NiFti files from axial 2D pngs.
• Code to separate synthesis training and test cases.
• Code to compute the image quality metrics such as SSIM, MSE, LPIPS, and more.
• Code to compute the Frèchet Inception Distance (FID) based on ImageNet and RadImageNet.
• The Duke Dataset used in this study is available on The Cancer Imaging Archive (TCIA).

Segmentation Code

• Code to prepare 3D single breast cases for nnunet segmentation.
• Train-test-splits of the segmentation dataset.
• Script to run the full nnunet pipeline on the Duke dataset.

Run the model

Model weights are stored on on Zenodo and made available via the medigan library.

To create your own post-contrast data, simply run:

pip install medigan

# import medigan and initialize Generators
from medigan import Generators
generators = Generators()

# generate 10 samples with model 23 (00023_PIX2PIXHD_BREAST_DCEMRI). 
# Also, auto-install required model dependencies.
generators.generate(model_id='00023_PIX2PIXHD_BREAST_DCEMRI', num_samples=10, install_dependencies=True)

Reference

Please consider citing our work if you found it useful:

@article{osuala2024simulating,
  title={Simulating Dynamic Tumor Contrast Enhancement in Breast MRI using Conditional Generative Adversarial Networks},
  author={Osuala, Richard and Joshi, Smriti and Tsirikoglou, Apostolia and Garrucho, Lidia and Pinaya, Walter HL and Lang, Daniel M and Schnabel, Julia A and Diaz, Oliver and Lekadir, Karim},
  journal={arXiv preprint arXiv:2409.18872},
  year={2024}
}

Acknowledgements

This repository borrows code from the pre-post-synthesis, the pix2pixHD and the nnUNet repositories. The 254 tumour segmentation masks used in this study were provided by Caballo et al.