Cite Synaptus (DOI):

License:

Available through Mathworks File Exchange:

Summary

Synaptus is a Matlab/Octave and Python toolbox for synthetic aperture or array imaging (NOTE: The Python version of Synaptus is work in progress). It was originally developed for ultrasonic imaging for non-destructive testing, but can be applied for similar imaging modes (e.g. ground penetrating radar). The toolbox focuses on algorithms implemented in the Fourier domain, and on imaging in multilayered structures (e.g. water, metal, rock).

The core functionality of the toolbox is to create focused images from raw (unfocused) pulse-echo data. Such data is produced by a transducer that transmits waves into a propagating medium, and records backscattered waves from within the medium. A backscattered "echo" is created when the waves interact with an object or layer with different physical properties than the propagating medium, e.g. a metal object in water. A measurement at a single point in space thus produces a 1-dimensional "depth profile". By moving the transducer laterally relative to the object under study, it is possible to create a 2- or 3-dimensional image of the object. A similar measurement can be performed using an array of multiple transducers. Due to the divergence of the transducer beams, the echoes from scattering objects are "smeared" laterally, making the images unfocused and hard to interpret. Examples of such images are given in the "Example raw and focused images" section below.

The algorithms in the Synaptus toolbox take such raw, unfocused images as input and processes them to create focused images. Conceptually, this is done by treating the raw images as measurements of a wave field, and using the wave equation to manipulate the wave field into a focused image. The details of this process is described in the PhD thesis included in the toolbox.

Background

The toolbox was originally written by Martin H. Skjelvareid, as a collection of algorithms developed during his work as a PhD candidate. The PhD thesis is included in the "docs" folder, and can also be downloaded from https://hdl.handle.net/10037/4649 . As with so many things, "the devil is in the details" when it comes practical implementation of synthetic aperture algorithms. The toolbox is meant to help people who are new to the field and are looking to make implementations of published algorithms. The data sets included in the toolbox will hopefully be useful in the development of new algorithms for similar measurement geometries. The toolbox could also represent a collection of reference methods against which new algorithms are compared.

The name of the toolbox is an abbreviation of "Synthetic APerTure UltraSound".

Example raw and focused images

The following images are taken from the thesis to illustrate some of the applications of the algorithms in the toolbox. The original figure numbering and captions have been included for context. See the thesis.pdf file under the 'docs' folder for further details.

Bottom drilled holes in PMMA and aluminium blocks

<img src="graphics/AcrylicGlassAndAluminiumLayers_Setup.png" width="600"> <img src="graphics/AcrylicGlassAndAluminiumLayers_3DRender.png" width="550"> <img src="graphics/AcrylicGlassAndAluminiumLayers_RawData.png" width="550"> <img src="graphics/AcrylicGlassAndAluminiumLayers_Focused.png" width="650">

Objects placed on cylindrical surface

<img src="graphics/ObjectsInPipe_Setup.png" width="550"> <img src="graphics/ObjectsInPipe_Images.png" width="600">

Rusted pipe imaged from inside

<img src="graphics/RustedPipeImaging_Setup.png" width="600"> <img src="graphics/RustedPipeInterior.png" width="600"> <img src="graphics/RustedPipe_Slices_RawAndFocused.png" width="600">

Algorithms

The main focus of the algorithms is on Fourier-domain synthetic aperture processing of ultrasound data. Fourier-domain processing is very common for synthetic aperture radar and sonar, but in the field of ultrasound, the time-domain "delay-and-sum" approach still dominates. With Fourier-domain processing, it is possible to extrapolate a sampled pulse-echo wavefield in both space and time ("wavefield migration"). One major advantage of this approach is that the wavefield is easily extrapolated between media with different wave velocities, enabling multi-layer imaging (very relevant for immersion ultrasound imaging).

The phase shift migration (PSM) algorithm works by migrating the recorded wavefield in small steps, and creating a focused image line/plane at each depth (using the "exploding reflector model"). From a processing standpoint, the method is not optimal, as the complete wavefield spectrum matrix has to be multiplied with a phase factor matrix at each step. However, since matrix multiplication is very fast in Matlab/Octave, the method is quite fast in practice.

The MULOK algorithm is a multi-layer version of the method known as Stolt migration, omega-k focusing, f-k focusing, etc. (several names exist for the same method). The wavefield extrapolation used in PSM is also used in MULOK, to extrapolate the wavefield between layers with different wave velocities. However, each layer is focused "all at once" by a resampling of the wavefield spectrum. This is generally a quite efficient method, but the practical performance very much depends on the efficiency of the interpolation method used (the linear interpolation mode of the interp1 function in Octave/Matlab is only moderately fast).

The CPSM algorithm is an adaptation of the PSM algorithm to a cylindrical imaging geometry (transducer pointing outward from cylindrical scanning surface). The solutions to the wave equation in a cylindrical geometry are Hankel functions. These can be used to extrapolate the wavefield, but are generally very time-consuming to compute. The algorithm also includes two alternative transfer functions which are approximate but much faster.

Organization

Assets common to both Python and Matlab versions

• datasets contains datasets in .mat-format, used for test/demonstration of the algorithms
• docs contains background information not specific to the programming language (PhD thesis ++)
• paper contains files realted to paper published in Journal of Open Source Software.

Python

The Python version of the toolbox is placed in the python folder and contains

• docs contains files related to documentation of the Python version (MkDocs).
• examples contains Jupyter notebooks demonstrating Python classes and functions on eaxmple datasets.
• learn contains Jupyter notebooks demonstrating simplified versions of some of the algorithms in the toolbox, with additional plots of data at intermediate steps to help understanding.
• synaptus contains the Python module(s) for synthetic aperture focusing
• test contains test code (pytest)

Matlab / Octave

The Matlab version of the toolbox is placed in the matlab folder and contains:

• core contains the functions for synthetic aperture focusing. Each file represents a separate algorithm.
• learn contains simplified versions of (some of) the algorithms in the toolbox, with additional plots of data at intermediate steps to help understanding.
• misc contains various functions used to help in processing and plotting of results.
• test contains test scripts for the algorithms

Requirements

Python

The Python version of synaptus depends on NumPy, SciPy, MatPlotLib, and Jupyter notebooks. See pyproject.toml for details.

Matlab / Octave

Using the Matlab/Octave part of synaptus requires a working base installation of Matlab or Octave, with some additional signal processing functions. Details for Matlab and Octave are given below.

Matlab

Matlab is available for Linux, MacOS and Windows, and can be purchased from Mathworks. Running Synaptus on Matlab requires:

• Base installation of Matlab (has been tested with version 2021a, installed on Windows)
• Signal processing toolbox (has been tested with version 8.6)

Octave

GNU Octave is free and open source, and is available for both Linux, MacOS and Windows. See GNU download and installation page for details.

Running Synaptus on Octave requires:

• Base installation of Octave (has been tested with version 5.2.0, installed on Ubuntu 20.04). On Debian / Ubuntu systems, Octave can be installed with sudo apt-get install octave.
• The signal package (has been tested with version 1.4.1). Octave installers for Windows bundle the signal package with the main application. On Debian/Ubuntu systems, the package can be installed using sudo apt-get install octave-signal Note that the signal package needs to be "loaded" inside Octave using the command pkg load signal. This can be automated by adding the command to a Octave startup file.

Installation

Python

Download the code from the synaptus repository, navigate to <synaptus_root_folder>/python, create a virtual environment for synaptus (optional but recommended)