RF-ULM: Radio-Frequency Ultrasound Localization Microscopy

Overview

NMS: Non-Maximum-Suppression
Map: Geometric point transformation from RF to B-mode coordinate space

SG-SPCN Architecture

Demos

1. ULM Animation Demo

rfulm_anim.mp4

Note: The video starts in slow motion and then exponentially increases the frame rate for better visualization.

2. Prediction Frames Demo

rfulm_rat18_short_clip.mp4

Note: Colors represent localizations from each plane wave emission angle.

Datasets

In vivo (inference): https://doi.org/10.5281/zenodo.7883227

In silico (training+inference): https://doi.org/10.5281/zenodo.4343435

Short presentation at IUS 2023

Installation

It is recommended to use a UNIX-based system for development. For installation, run (or work along) the following bash script:

> bash install.sh

Note that the dataloader module is missing in this repository. My implementation is a hacky version of the work found at https://github.com/AChavignon/PALA, which was used as a reference in this project. When using data other than mentioned here, one would need to start writing this part from scratch. The simpletracker repository has not been used in the TMI publication and can be ignored.

Citation

If you use this project for your work, please cite:

@article{hahne:2024:rfulm,
  author={Hahne, Christopher and Chabouh, Georges and Chavignon, Arthur and Couture, Olivier and Sznitman, Raphael},
  journal={IEEE Transactions on Medical Imaging}, 
  title={RF-ULM: Ultrasound Localization Microscopy Learned From Radio-Frequency Wavefronts}, 
  year={2024},
  volume={43},
  number={9},
  pages={3253-3262},
  keywords={Location awareness;Radio frequency;Array signal processing;Superresolution;Convolution;Ultrasonic imaging;Kernel;Super-resolution;ultrasound;localization;microscopy;deep learning;neural network;beamforming},
  doi={10.1109/TMI.2024.3391297}
}

Acknowledgment

This research is funded by the Hasler Foundation under project number 22027.