Coding Mask Design for Single Sensor Ultrasound Imaging

Abstract

We study the design of a coding mask for pulse-echo ultrasound imaging. We are interested in the scenario of a single receiving transducer with an aberrating layer, or 'mask', in front of the transducer's receive surface, with a separate co-located transmit transducer. The mask encodes spatial measurements into a single output signal, containing more information about a reflector's position than a transducer without a mask. The amount of information in such measurements is dependent on the mask geometry, which we propose to optimize using an image reconstruction mean square error (MSE) criterion. We approximate the physics involved to define a linear measurement model, which we use to find an expression for the image error covariance matrix. By discretizing the mask surface and defining a discrete number of mask thickness levels per point on its surface, we show how finding the best mask can be posed as a variation of a sensor selection problem. We propose a convex relaxation in combination with randomized rounding, as well as a greedy optimization algorithm to solve this problem. We show empirically that both algorithms come close to the global optimum. Our simulations further show that the optimized masks have better a MSE than nearly all randomly shaped masks. We observe that an optimized mask amplifies echoes coming from within the region of interest (ROI), and strongly reduces the correlation between echoes of pixels within the ROI.

Compressive Imaging with Spatial Coding Masks on Low Number of Elements: An Emulation Study

“…In this paper, only one randomly generated coding pattern was adopted, which might still be optimized towards our application scenario. Our previous work [6] has shown that this may improve results further. Furthermore, we are aiming to apply our device for flow/function monitoring in the carotid artery, for which the requirements of spatial separability might be more lenient than for B-mode imaging.…”

Multiple Measurement Vector Model for Sparsity-Based Vascular Ultrasound Imaging

“…Using the matched filter A, we can obtain the system in (2), Z = A H X = A H AS + N = HS + N, with H = A H A, which is not a shift-invariant matrix. Here, A is constructed using the k-wave toolbox in MATLAB for an imaging system with 128 transmitters and receivers [26] [27]. Furthermore, the A matrix is constructed in the frequency domain instead of the time domain.…”