Ultrasound Computed Tomography (USCT)

Ultrasound Computed Tomography

Ultrasound Computed Tomography (USCT) is an imaging technique that utilizes tomographic principles to obtain quantitative estimates of acoustic properties such as speed-of-sound, density, and acoustic attenuation. USCT has a number of potential applications, including breast cancer screening. It has a number of advantages over existing imaging modalities. Compared with mammography, it is radiation-free, breast-compression-free, and relatively inexpensive. Compared with conventional B-mode ultrasound, it produces images with a large field-of-view whose quality is independent of the skill of the operator.

Most image reconstruction methods in USCT are based on approximate solutions to the acoustic wave equation, which ignore higher-order diffraction effects. This can result in reconstructed images with poor resolution. This problem can be overcome through the use of waveform-inversion-based image reconstruction methods that directly seek the solution of the acoustic wave equation. However, this approach is often computationally very expensive. To circumvents this problem, we proposed the efficient algorithm refer as waveform inversion with source encoding (WISE). We also proposed the dual averaging method to improve the effectiveness of regularization.

Our work focuses on advanced image reconstruction algorithms and accurately modeling of the underlying physics of USCT imaging systems. We accomplish this through the use of optimization-based image reconstruction algorithms that leverage the latest work in stochastic optimization, machine learning, and computational acoustics. This project represents a collaboration with Dr. Neb Duric and his team.

(a) Sound speed distribution of the numerical breast phantom; (b) Images reconstructed by use of the WISE method; (c) the bent-ray model-based sound speed reconstruction method; (d) Profiles  of the reconstructed images

Sound speed images of a breast cancer patient reconstructed by use of ray-less adjoint state method(left), WISE method(middle) and weighted RDA method with a wavelet-based penalty and regularization parameter. (right)

Publication