GPU acceleration of acoustic radiation force impulse imaging using scanning-mode

Acoustic radiation force impulse (ARFI) system has a more simple hardware architecture than the supersonic shearwave imaging (SSI), which make it easier to be miniaturized for some special clinical practice, such as the emergency treatments in the field first-aid and the donor evaluation for organ transplantation. However, currently ARFI cannot provide 2D stiffness images like SSI can do. To fit this gap, the time-consuming algorithms of ARFI are to be migrated to the graphics processing unit (GPU) for substantial acceleration. In this study, the algorithms of ARFI were modified and improved to fit for the parallel computation on GPU: an analytic signal based cross-correlation method was used to assess the tissue displacements, a cyclic reduction (CR) method was used to calculate the cubic spline interpolation, and a time-of-flight (TOF) method based on Radon transformation was used to determine the shear wave velocity. The ratio of the time consumptions of the programs running on CPU and on GPU was calculated using the practical data collected by our self-developed ARFI system. For the most time-consuming parts such as cross-correlation, filtering, Radon transformation and cubic spline interpolation, large acceleration ratio has been confirmed between the two programs. And a quasi-real-time scanning-mode 2D ARFI (s2D-ARFI) imaging system was implemented by the help of GPU acceleration. Several imaging results on phantom, isolated animal tissue, and in vivo animal are also presented.