This is an example of ultrasound beamforming using a linear array in both Python and Rust. The Python script is written with something of an educational emphasis, and the Rust script is in a minimally-functional state to be refined as Rust's signal and image processing libraries mature.
RF Data was simulated using a 3rd party MATLAB toolbox called K-Wave. Specifically, the data was generated using example_us_bmode_linear_transducer.m, which sets up a linear probe and generates the signals received after pulsing into a 3D scattering phantom. The phantom accounts for nonlinearity, multiple scattering, power law acoustic absorption, and a finite beam width in the elevation direction. One can find the nature of the simulated data as well as a description of the K-wave program here.
The aforementioned m-file not only simulates recorded data but performs image reconstruction as well. However, I have merely acquired the raw RF data stored in the variable sensor_data and written my own image reconstruction routines. You know, for fun.
The conventional steps in an ultrasound signal processing pipeline are conducted, and a comparison is performed between simple B-mode imaging, beamforming with fixed receive focus, and dynamic focusing:
This program requires numpy, scipy, matplotlib, and h5py.
The available signal and image processing libraries written in pure Rust are in their infancy at the time of this writing, so this script makes use of C++ libraries such as fftw and opencv. The script is centered around ndarray and incorporates logging (simplelog) and plot creation/saving (plotlib) for the checking of intermediate outputs.
Currently a beamformed image is produced with dynamic focusing, albeit with a slow upsampling step and no signal filtering.
You may be required to install fortran/openblas to compile (e.g. sudo apt-get install gfortran libopenblas-dev on debian). Notes on installing a version of OpenCV compatible with rust bindings are included.
| py | rs |
|---|---|
 |
 |