GAFFA - Global Anatomical Feasibility Filter and Analysis

Overview

This is the official implementation of the paper:
"Implicit Is Not Enough: Explicitly Enforcing Anatomical Priors inside Landmark Localization Models"
Published in Bioengineering (DOI: 10.3390/bioengineering11090932)

Citation

If you find this work helpful, please cite our paper:

@Article{bioengineering11090932,
    AUTHOR    = {Joham, Simon Johannes and Hadzic, Arnela and Urschler, Martin},
    TITLE     = {Implicit Is Not Enough: Explicitly Enforcing Anatomical Priors inside Landmark Localization Models},
    JOURNAL   = {Bioengineering},
    VOLUME    = {11},
    YEAR      = {2024},
    NUMBER    = {9},
    ARTICLE-NUMBER = {932},
    PubMedID  = {39329674},
    ISSN      = {2306-5354},
    DOI       = {10.3390/bioengineering11090932}
}

License

This repo incorporates MedicalDataAugmentationTool and is therefore licensed under the GNU General Public License v3.0.

Installation

1. Dataset Setup

   • Place dataset images in datasets/xray_hand/images (raw+mhd files)
   • Annotations are included in the repository

2. Environment Setup

   # Create conda environment with Python 3.10.14
   conda create -n GAFFA python=3.10.14
   conda activate GAFFA

3. Dependencies

   pip install -r requirements.txt

Usage

Run the example configuration files using:

python main.py default/default_xray_hand_train_UNet+GAFFA.json

Important Notes

• Conditional distribution heatmaps must be precomputed locally (default setting in configs)
• This precomputation is required even when using pretrained GAFFA models (generates a .npy file containing landmark connectivity information)
• As a PhD student, repository maintenance will be sporadic, but I will attempt to address minor questions when possible