Connecto-Radiomics

A pipeline for radiomics informed connectomics.

Set up

There are two conda enironments needed to run the pipeline:

um2ii_rad_env: the environment that will be needed to run majority of the pipeline, including any of the inidiviudal scripts.
um2ii_pyrad: this environment is exclusivley for pyradiomics and running the radiomics calculations.

The radcon.sh driver script takes care of switching between environment as needed.

Usage

The main driver script is the radcon.sh bash script.

To execute: radcon.sh [settings].json See below for explanation of all parameters in the setting JSON file.

This script will do the following in order

Image segmentation using a trained model.
For each image, separate out the each segmentation into an individual nifti file.
Run the radiomics calculations for each individual segmentation using Pyradiomics, and clean the results
Run the connectomics and output individual graph-networks for each image.
Caclulate connectomics metrics and output all radiomics and connectomic calculations in a single csv file.

Generally speaking, there is a single python script responsible for each of the steps listed above in the /src/ directory. They can be ran individually if needed. See section below on Script-by-Script documentation.

Caveats:

radcon.sh makes the image segmentations using a trained nnUNET model and in nifti format. This line of code is emphasized in the bash script and may need to be modifed per use case.
This pipeline was built and tested using knee MRIs from the OAI dataset. This loosely affects the naming conventions, so there may be issues with name parsing,...etc. if applying to other datasets. These issues shouldn't be difficult to resolve. The expected name of the input images is: [pt_id] _ [time_point] _ [mri_id] _ [image_id] _ 0000.nii.gz .

Note: Reference the test_run folder for example output and the parameters.json file for example settings.

Settings JSON file

Parameter	Description
segmentation_separation
source_dir	path to raw images in nifti format
dest_dir	path where segmentations should be stored
segmentation_nifit_seg_path	path where nifti files for each individual segmentation should be stored
num_segmentations	the number of segmentations the model predicts

radiomics
dir	directory where all radiomics files are stored
thread_count	number of threads to use for

connectomics
dir	directory to store all connectomics files
two_tailed_filtering_level	threshold to filter off edge weights for the graphs. (i.e. a value of 15 would mean that only the top and bottom 15% of edge weights will be kept.)

feature_dict	a dictionary that holds the anatomical label for each label id

Script by script description