Adaptive Data Function for Robust Ultrasound Elastography

Abstract

Regularized optimization-based ultrasound elastography techniques minimize an energy function consisting of data and continuity terms to obtain the displacement tensor between radio-frequency (RF) frames. The data term associated with the existing energy-based techniques takes only the amplitude similarity into account and hence lacks robustness to the outlier samples present in the RF frames. This drawback creates noticeable artifacts in the strain image. To address this issue, we devise the data function as a linear combination of the amplitude and gradient residuals. We follow an iterative scheme to estimate the adaptive weight associated with each similarity term. Finally, we convert the non-linear optimization problem to a sparse system of linear equations which is solved for millions of variables in an efficient manner. We name our technique rGLUE: robust data term in GLobal Ultrasound Elastography. We validate rGLUE using simulation and in vivo breast datasets. In both of the experiments, rGLUE proves its robustness to outliers and outperforms state-of-the-art time-delay estimation technique both visually and quantitatively. For the noisy simulation data, the proposed rGLUE technique improves the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) from 1.67 to 7.04 and 2.89 to 13.46, respectively. In case of the breast datasets, SNR and CNR improves from 6.35 to 7.81 and 7.94 to 9.90, respectively.