Stability analysis of dynamic radiomics features for cardiac magnetic resonance

Introduction

This repository contains the source code for the paper

Stability of Dynamic Radiomics Features in Cardiac MRI by Mike D. Klaus, Fabian Laqua, Bettina Baeßler, and Markus J. Ankenbrand. medRxiv 2025.04.18.25326051

General workflow

Data sets

The large data in this repository is managed via dvc. That's why you see a lot of .dvc files in the folder structure. The content of the corresponding files is not directly included. For many of the MRXCAT and ACDC files and all the final results, these files are included in the dvcstore deposited at Zenodo. In order to use it, download and unzip it into your repository root. Then you can get files from the store with dvc pull -r zenodo. For example:

dvc pull -r zenodo analysis/features_normalized/ACDC/patient001.csv.dvc

MRXCAT

Simulations with different SNR (5,10,20,30) using MRXCAT [1] version v1.4 and the breathhold phantom. Code is in code/mrxcat_simulations and simulation results in data/mrxcat_simulations. The simulations were performed with Matlab version R2021b. Simulation results have to be converted from the cpx format to the nifti format for further analysis. Steps include:

1. .cpx to .csv convertion
2. .csv to .nifti convertion The masks were then extracted from the breathhold phantom used for the simulation in MRXCAT.

ACDC

The ACDC dataset was published as part of the Automatic Cardiac Detection Challenge [2].

BAE

Data from the publication A systematic evaluation of three different cardiac T2-mapping sequences at 1.5 and 3T in healthy volunteers [3].

Running the analysis

The analysis is implemented as a snakemake workflow (see Snakemake and .smk files in workflow/). If you want to re-run it, you can call this command from the root of the repository:

snakemake --software-deployment-method conda --cores 8

The default target are all the figures and tables in figures/ (including supplementary ones). So this command will likely complain about missing input files (the BAE data). Either edit the inputs for rule all manually in workflow/Snakefile or call snakemake with the outputs of interest, e.g.:

snakemake --software-deployment-method conda --cores 8 figures/fig1.svg

References

1. Wissmann, L., Santelli, C., Segars, W.P. et al. MRXCAT: Realistic numerical phantoms for cardiovascular magnetic resonance. J Cardiovasc Magn Reson 16, 63 (2014). https://doi.org/10.1186/s12968-014-0063-3
2. Bernard, O., Lalande, A., Zotti, C., Cervenansky, F., et al. Deep Learning Techniques for Automatic MRI Cardiac Multi-structures Segmentation and Diagnosis: Is the Problem Solved? IEEE Transactions on Medical Imaging, 37, 11, 2514-2525 (2018). https://doi.org/10.1109/TMI.2018.2837502
3. Baeßler, B. et al. A systematic evaluation of three different cardiac T2-mapping sequences at 1.5 and 3T in healthy volunteers. Eur. J. Radiol. 84, 2161–2170 (2015). https://doi.org/10.1016/j.ejrad.2015.08.002