Randomized transmit and receive ultrasound tomography

TL;DR

This paper introduces a randomized ultrasound tomography method that uses specific source and receiver signal randomization to produce well-conditioned matrices, enabling effective image reconstruction even with low SNR and without traditional approximations.

Contribution

The paper presents a novel randomization technique for ultrasound tomography that improves matrix conditioning and image quality in noisy environments, surpassing traditional methods.

Findings

Successful image reconstruction at SNR as low as 1

Randomization yields well-conditioned matrices for inversion

Superiority over synthetic aperture in low SNR conditions

Abstract

A tomographic method is considered that forms images from sets of spatially randomized source signals and receiver sensitivities. The method is designed to allow image reconstruction for an extended number of transmitters and receivers in the presence noise and without plane wave approximation or otherwise approximation on the size or regularity of source and receiver functions. An overdetermined set of functions are formed from the Hadamard product between a Gaussian function and a uniformly distributed random number set. It is shown that this particular type of randomization tends to produce well-conditioned matrices whose pseudoinverses may be determined without implementing relaxation methods. When the inverted sets are applied to simulated first-order scattering from a Shepp-Logan phantom, successful image reconstructions are achieved for signal-to-noise ratios (SNR) as low as 1. Evaluation of the randomization approach is conducted by comparing condition numbers with other forms of signal randomization. Image quality resulting from tomographic reconstructions is then compared with an idealized synthetic aperture approach, which is subjected to a comparable SNR. By root-mean-square-difference comparisons it is concluded that - provided a sufficient level of oversampling - the dynamic transmit and dynamic receive approach produces superior images, particularly in the presence of low SNR.