Inverse problems of experimental data interpretation in 3D ultrasound tomography

The inverse problem of 3D ultrasound tomography is considered in this paper as a nonlinear coefficient inverse problem for a hyperbolic equation. The employed mathematical model accurately describes the effects of ultrasound wave diffraction and absorption in inhomogeneous media. The velocity of acoustic waves inside the test sample is reconstructed as an unknown function of three spatial coordinates. The number of unknowns in the nonlinear inverse problem reaches 50 million. The developed iterative algorithms for solving the inverse problem are designed for GPU clusters. The main result of this study is testing the developed algorithms on experimental data. The experiments were carried out using a 3D ultrasound tomographic setup developed at Lomonosov Moscow State University. Acoustic properties of the test samples were close to those of human soft tissues. The volume of data collected in experiments is up to 3 GB. Experimental results show the efficiency of the proposed algorithms and confirm that the mathematical model is adequate to reality. The proposed algorithms were tested on the GPU partition of Lomonosov-2 supercomputer.