mar-eval is the reference implementation for objectively evaluating Metal Artifact Reduction (MAR) algorithms according to the proposed Annex GG of IEC 60601-2-44 Ed.4.

Unlike previous versions, this toolkit implements DDOG (Dense Difference of Gaussian) channels and provides a universal .npz data interface, making it compatible with DukeSim, CatSim, and manufacturer-internal ray tracers.

• Regulatory Ready: Auto-generates the $\Delta AUC$ statistical report and detectability curves required for FDA 510(k) submissions.
• Simulator Agnostic: Accepts data from any source via the .npz container format.
• Interactive CLI: A guided workflow for simulation or user-data processing.
• Physics Engine: Includes an analytical simulation mode to generate reference curves based on Table GG.1.
• Educational Demo: Interactive Jupyter Notebook explaining the math and physics of Annex GG—runnable in-browser via Binder.

git clone https://github.com/cdc15000/mar-eval.git
cd mar-eval
pip install -r requirements.txt

Click the Launch Binder badge at the top of this README to run the analysis engine in your browser. This demonstrates the DDOG channel filters and the AUC detectability physics.

python src/mar_eval.py

Select Option 1 (Simulation) to see the reference detectability curves immediately.

1. Generate: Run your simulations (11 Doses × 7 Contrasts).
2. Index: Run python tools/generate_manifest.py to index your DICOM files.
3. Pack: Run python tools/dicom_converter.py to create manufacturer_data.npz.
4. Analyze: Run src/mar_eval.py and select Option 2 (Upload).

• Regulatory Guide: How to demonstrate Substantial Equivalence (Head, Chest, Abdomen).
• Simulator Guide: How to configure DukeSim/CatSim for the Table GG.1 batch.
• Contributing: Guidelines for researchers and engineers wishing to improve the toolkit.

If you use this tool in regulatory submissions or research, please cite:

C.D. Cocchiaraley, Annex GG — Objective evaluation of Metal Artifact Reduction algorithms in CT imaging, Proposed addition to IEC 60601-2-44 Ed. 4 (2025).