Data

Tensorflow implementation of optimizing a neural representation for a multiple ultrasound sweeps for the same region of interest.

Ultra-NeRF: Neural Radiance Fields for Ultrasound Imaging
Magdalena Wysocki*¹, Mohammad Farid Azampour*¹,², Christine Eilers¹, Benjamin Busam¹,³, Mehrdad Salehi¹, Nassir Navab^1
1Technical University of Munich (TUM), ²Sharif University of Technology, ³3Dwe.ai
*denotes equal contribution
in MIDL 2023 (Oral Presentation)

To setup a conda environment, download example training data, begin the training process, and launch Tensorboard:

conda env create -f environment.yml
conda activate ultra_nerf

Note Please, contact me in case of any issues with running the code.

python run_ultra_nerf.py --config conf_us.txt --expname test_generated --n_iters 200000 --loss ssim --i_embed_gauss 0 --i_img 2000 --i_print 2000  --i_weights 2000

Our data consist of several sweeps of the same region of interest taken from different observation angles (a). The poses are calibrated.

Synthetic data (c): 0.31x 0.27 mm, depth 140 mm, width 80 mm

Phantom data (b): 0.22 x.07 mm, depth 100 mm, width 38 mm

The image shows the coordinate system and sampling method (equidistant sampling).

Added for the synthetic dataset. For the phantom dataset will soon be updated.

Sample results

@inproceedings{wysocki2023ultranerf,
  title={Ultra-NeRF: Neural Radiance Fields for Ultrasound Imaging},
  author={Magdalena Wysocki and Mohammad Farid Azampour and Christine Eilers and Benjamin Busam and Mehrdad Salehi and Nassir Navab},
  year={2023},
  booktitle={MIDL},
}

ultra-nerf is available under the MIT License. For more details see: LICENSE and ACKNOWLEDGEMENTS.

Large parts of the code are from the tensorboard NeRF implementation. See ACKNOWLEDGEMENTS.