kwahid / C3RO_demographics_analysis

For C3RO secondary analysis looking at demographic variable correlation to segmentation quality. Collaboration between Fuller lab at MDA and Gillespie lab at MSK.

Determining The Role Of Radiation Oncologist Demographic Factors On Segmentation Quality: Insights From A Crowd-Sourced Challenge Using Bayesian Estimation

Repo for code related to C3RO demographics analysis project.

Pre-print available at: https://www.medrxiv.org/content/10.1101/2023.08.30.23294786v2. Corresponding image sets used for this project are avaliable on Figshare (data doi: 10.6084/m9.figshare.21074182). Data descriptor located at: https://www.medrxiv.org/content/10.1101/2022.10.05.22280672v1. All data has been anonymized to remove patient PHI. The CSV files generated as data for this specific project can also be found on Figshare (doi:10.6084/m9.figshare.24021591).

This repo contains the following files:

Jupyter notebook of scripts to generate CSV files for regression analysis (Base_file_generation.ipynb).

Jupyter notebook of scripts to perform statistical analysis in paper (Statistical_analysis_code.ipynb).

Folders for various output files (bambi_binary_bayesian_regression_outputs, csv_files, descriptive_stats, plot_outputs).

Conda virtual Environment reproduction:

For ease of code reproduction, we have generated a YAML file with the conda virtual environment parameters that were used to run the most recent version of the code (C3RO_regression.yml). We used conda version 4.12.0. Unfortunately, some of the conda packages we used implemented Windows-specific build hash strings, so exact decimal-level reproducibility precision cannot be ensured if you try to install the environment on a different OS. Also please note, one of the required libraries for metric calculations (surface_distance) will require the user to pull a GitHub repo onto their local folders. See instructions here:https://github.com/google-deepmind/surface-distance.

To install a new conda environment on your system and ensure you can access the kernel in Jupyter, use the following commands:

1. conda env create -f C3RO_regression.yml -n C3RO_regression
2. python -m ipython install --user --name C3RO_regression --display-name "C3RO_regression"

Utilized the following core Python (version 3.11.4) libraries in project:

arviz version 0.15.1.
bambi version 0.12.0.
pymc version 5.6.1.
SimpleITK version 2.3.0.
Numpy version 1.24.4.
Pandas version 2.0.3.
Surface-Distance-Based-Measures version 0.1.
Matplotlib version 3.7.2.
Seaborn version 0.12.2.

Collaboration between Fuller lab at MDA and Gillespie lab at MSK. For more information on the Fuller lab and associated projects please visit: https://www.mdanderson.org/research/departments-labs-institutes/labs/fuller-laboratory.html.