nyukat / mri_ai

This repository contains code that was used to train and evaluate deep learning models, as described in the article "Improving breast cancer diagnostics with artificial intelligence for MRI" by Jan Witowski et al.

Improving breast cancer diagnostics with artificial intelligence for MRI

Introduction

This repository contains code that was used to train and evaluate deep learning models, as described in the article "Improving breast cancer diagnostics with artificial intelligence for MRI" by Jan Witowski et al. It includes model code, training loop, utilities, Jupyter notebooks and scripts used to evaluate results and generate plots.

TCGA-BRCA processing

This repository contains script for processing the TCGA-BRCA breast MRI data set. Specifically, there are two key files in the repository:

tcga_brca/tcga_brca_key.yaml

This file identifies all pre- and post-contrast sequences for all examinations in the data set. For example, 102: t1pre indicates that series number 102 contains pre-contrast images. Additionally, this file recognizes different ways in which volumes are saved. In some MRI exams, left and right breasts are saved in separate series. Please refer to the file for more details.

tcga_brca/Process_TCGA_BRCA.ipynb

This jupyter notebook utilizes information from the tcga_brca_key.yaml to do the following tasks:

• load DCE-MRI exams from the TCGA-BRCA data set,
• identify pre- and post-contrast sequences,
• resample them to the same anisotropic spacing,
• reorient them to LPS orientation,
• save them to a standardized nifti (.nii) format, so that for each breast MRI exam there are three files: t1pre.nii.gz, t1c1.nii.gz, and t1c2.nii.gz.

Reference

@article{witowski2022improving,
   title = {Improving breast cancer diagnostics with artificial intelligence for MRI}, 
   author = {Jan Witowski and Laura Heacock and Beatriu Reig and Stella K. Kang and Alana Lewin and Kristine Pysarenko and Shalin Patel and Naziya Samreen and Wojciech Rudnicki and Elżbieta Łuczyńska and Tadeusz Popiela and Linda Moy and Krzysztof J. Geras},
   journal = {medRxiv:10.1101/2022.02.07.22270518}, 
   year = {2022}
}