XRA-Diff: Guided diffusion as a refinement for backprojected coronary angiography X-ray images

If you find any of these useful please cite our article (see: reference.bib)

Installing dependencies

uv sync
uv pip install -e .

Preparing data

uv run scripts/dataset/prepare_data.py
uv run scrpits/dataset/make_split.py

Scripts

Under scripts directory:

• experiments/ - scripts for running experiments
  • synthetic/ - synthetic dataset tests
  • clinical/ - clinical dataset tests
  • ablation/ - ablation tests
  • other/
    • test_synthetic_motion_right.py - add simulated motion to synthetic dataset
• demos/ - scripts for qualitative evaluation
  • synthetic_visualizer.py - synthetic data visualizer (see --help)
  • clinical_visualizer.py - clinical data visualizer (see --help)
• dataset/ - dataset preparation
  • image2cas.sh - extraction of ImageCAS
  • prepare_data.py - simulates the image acquisition using cone beam geometry, generates synthetic dataset
  • make_split.py - splits the above dataset to train and valitation splits
  • make_pilot.py - create a validation dataset subset for pilot study
• training/ - scripts for training models
  • train_diffusion_right.py - train diffusion model for right arteries
  • train_diffusion_left.py - train diffusion model for left arteries
  • train_gan_right.py - train gan model for right arteries (baseline)
  • train_gan_left.py - train gan model for left arteries (baseline)
  • train_unet3d_right.py - train unet for right arteries (baseline)
  • train_unet3d_left.py - train unet for left arteries

Notebooks

Under notebooks directory:

• synthetic_results.ipynb (3.1. State-of-the-art)
• motion.ipynb (3.2. Dataset domain gap)
• clinical_results.ipynb (3.3. Clinical feasibility)
• ablation.ipynb (3.4. Ablation study)
• clinical_qualitative.ipynb (3.5. Qualitative analysis)

Model weights

Are available at Zenodo under 17413997

Experiment results

CSV's with our measurements are also at Zenodo under 17413997

Some examples