Reliable Radiomics

This is the code for the paper "Reliability as a Precondition for Trust –Segmentation Reliability Analysis of Radiomic Features Improves Survival Prediction"

If you use the results or code, please cite the following paper:

@article{
}

Results

The ICC(1) value for each dataset and feature can be found in reliability.json (median with 95%-confidence interval) or in reliability.csv (only median).

Run with your own data

Step 1: Install

cd .../ReliableRadiomics/
pip install numpy 
pip install . 

Step 2: Data (/data)

• Put your segmented images under /data/images/$DatasetName/images.hdf
• Put your radiomics settings under /data/settings/
• Put your survival data under /data/images/$DatasetName/survival.csv

Make sure your images.hdf has the following keys and shape:

• uids ~ [Samples, ] with PatientID_LesionIdx_SliceIdx
• images ~ [Samples, z,y,x]
• labels ~ [Samples, Annotations, z,y,x]
• spacing ~ [Samples , 3] with order (x,y,z)

Step 3: Execute (/scripts)

1. main_radiomics_computation.py
   • Script for calculating the Radiomic features. Note that features are only calculated from segmentation that exceeds the dice_min threshold.
2. main_radiomics_evaluation.py
   • Script for calculating and visualizing the inter-rater reliability (ICC scores) from the Radiomic features.
3. main_radiomics_evaluation_between.py
   • Script to calculate and visualize the inter-rater reliability (ICC scores) of the Radiomic features across multiple datasets.
4. main_survival_computation.py
   • Script to estimate overall survival using a Cox model.
5. main_survival_evaluation.py
   • Script to visualize the variance of survival predictions as a function of inter-rater reliability.

• Note: Some minor code changes (setting the correct $DatasetName) at the beginning of each .py file may be required to use your specific dataset