The architecture of specialized GPU clusters used for solving the inverse problems of 3D low-frequency ultrasonic tomography

Abstract

This paper is dedicated to the development of the architecture of specialized GPU clusters that can be used as computing systems in medical ultrasound tomographic facilities that are currently being developed. The inverse problem of ultrasonic tomography is formulated as a coefficient inverse problem for a hyperbolic equation. An approximate solution is constructed using an iterative process of minimizing the residual functional between the measured and simulated wave fields. The algorithms used to solve the inverse problem are optimized for a GPU. The requirements for the architecture of a GPU cluster are formulated. The proposed architecture accelerates the reconstruction of ultrasonic tomographic images by 1000 times compared to what is achieved by a personal computer.