Fast image acquisition in pulse-echo ultrasound imaging using compressed sensing

Abstract

For the plane-wave imaging challenge in medical ultrasound (PICMUS) data sets, we applied an enhanced version of our method for fast image acquisition in pulse-echo ultrasound imaging (UI) based on compressed sensing (CS). Our method, which is based on the theory provided in [1]–[8], formulates the ultrafast pulse-echo imaging process as a linear inverse scattering problem (ISP). The ISP is derived from a realistic physical model for the propagation, absorption, and scattering of ultrasonic waves in biological soft tissues. The ISP recovers fluctuations in the tissues' compressibility from observations of the scattered sound field. For the sequential emission of only a few steered plane waves (PWs), the ill-posed and underdetermined ISP is regularized based on CS. We postulate the existence of a sparse representation of the fluctuations to be recovered under a known linear transform, e.g. the Canonical basis or wavelets. Our method is implemented efficiently in the temporal frequency domain using a version of the fast multipole method parallelized on a graphics processing unit. Emitting only a single steered PW, the mean {axial; lateral} widths were {90 µm; 110 µm} for the simulation data and {330 µm; 370 µm} for the experimental data, corresponding to improvements up to 85%compared to the reference delay-and-sum algorithm provided by the organizers. The mean contrast metrics were 14.51 dB for the simulation data and 8.7 dB for the experimental data, corresponding to improvements up to 46% compared to the reference algorithm. The sequential emission of three steered PWs resulted in the corresponding widths and contrast metrics of {80 µm; 110 µm}, {370 µm; 330 µm}, 14.57 dB, and 6.35 dB, respectively.