Differentiable Beamforming for Ultrasound Autofocusing

Project Page | Paper | Pre-Print | Data

Walter Simson, Louise Zhuang, Sergio Sanabria, Neha Antil, Jeremy Dahl, Dongwoon Hyun
Stanford University

This is the official implementation of the paper "Differentiable Beamforming for Ultrasound Autofocusing."

High-Level Overview

• dbua.py - main experiment file. Adjust the global configuration parameters to run experiments.
• das.py - Delay-and-sum IQ data according to a given time delay profile.
• paths.py - Calculates the time-of-flight between two points according to a speed-of-sound map.
• losses.py - Contains the proposed phase-error and auxiliary loss functions.

Getting Started

This project requires ffmpeg to be installed to save mp4 files. If you have not already, install ffmpeg on Ubuntu by running:

sudo apt-get install ffmpeg

You can install the required Python dependencies like this:

micromamba env create -f environment.yml
micromamba activate dbua

Then, install JAX using the instructions for the pip GPU installation method found here.

Reproducing results

The hyperparameters set in dbua can be used to reproduce the results from the paper. The data from the paper can be found in the release on GitHub, and the mat files should be placed in the data/ directory of this repository:

data
├── 1420.mat
├── 1465.mat
├── 1480.mat
├── 1510.mat
├── 1540.mat
├── 1555.mat
├── 1570.mat
├── checker2.mat
├── checker8.mat
├── four_layer.mat
├── inclusion_layer.mat
├── inclusion.mat
├── README.md
└── two_layer.mat

Run the program with this command:

python dbua.py

Citation

@inproceedings{simson2023dbua,
            title={Differentiable Beamforming for Ultrasound Autofocusing},
            author={Simson, Walter and Zhuang, Louise and Sanabria, Sergio J and Antil, Neha and Dahl, Jeremy J and Hyun, Dongwoon},
            booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
            pages={428--437},
            year={2023},
            organization={Springer}
          }

FAQ:

• Q: What computer configuration was used to develop this program?
• A: This code was developed using the following configuration:

Attribute	Detail
OS	Ubuntu Linux
RAM	32GB
GPU	NVIDIA RTX A6000 (48 GB VRAM)
CUDA Version	12.1