Scripts

This document outlines the utility scripts available in the repository.

compute_metrics.py
Evaluates segmentation results by computing Dice score, Hausdorff distance, and surface distances.
Outputs patient_wise_metrics.csv and aggredated evaluation_results.csv.
convert_TCIA_to_nnunet.py
Converts the TCIA pediatric dataset into the nnU-Net compliant format.
create_totalseg_subset.py
Creates a balanced subset of the TotalSegmentator dataset for fingerprinting (P_m) on an equal number of pediatric and adult cases.
remap_labels.py
Remaps segmentation labels to adhere to our unified labeling scheme.
run_TotalSegmentator.sh
Executes the TotalSegmentator pipeline for baseline inference on various test sets.
get_results.py
Processes segmentation metrics from multiple models across different datasets, performs statistical comparisons against baseline models, identifies the best-performing scores per region of interest, and outputs the results as a formatted LaTeX table.

Installation

To evaluate segmentation results using compute_metrics.py, you need to install the surface-distance package. You can install it via pip:

$ git clone https://github.com/deepmind/surface-distance.git
$ pip install surface-distance/

Once installed you can run, for example:

python scripts/compute_metrics.py reference_dir prediction_dir

This will produce patient_wise_metrics.csv and an aggregated evaluation_results.csv in the current directory.