A Comparative Study of Displacement De-Noising Strategies: An in Vivo Feasibility Study Using 3D Whole Breast Ultrasound Data

With the availability of 3D whole breast ultrasound data (WBUS), a lot of recent research efforts are devoted to 3D ultrasound strain elastography. We propose a software post-processing strategy that can be used to further improve ultrasonically-tracked displacements, particularly, off-axis components of ultrasonically-tracked displacements. A correlation-based speckle tracking algorithm was first used to obtain relatively accurate 3D displacement fields. Then, three different denoising algorithms were applied and evaluated. The first denoising algorithm is a partial differential equation (PDE)-based algorithm and tissue incompressibility was enforced through iterative optimization. The second method utilizes radial basis function (RBF) also to enforce tissue incompressibility, while the third method is a non-local median filter method. Based on data investigated, the PDE-based algorithm achieved the lowest incompressibility among three different algorithms. As compared to the raw displacement estimates, reduction rates of incompressibility were approximately 95%, 50% and 63% after three denoising algorithms had been applied, respectively. The analysis of 3D ultrasound data acquired using a clinical scanner (Acuson S2000 ABVS, Siemens Medical Solution (USA) Inc., Mountain View, CA) demonstrated that the PDE-based denoising algorithm can provide most consistent lateral strain images. In conclusion, our preliminary results demonstrated that denoising can significantly improve non-axial displacement estimates.