A multi-GPU acceleration for 3D imaging of the prostate

Transrectal Electric Impedance Tomography (TREIT) has been proposed jointly with ultrasound (US) imaging of the prostate to enhance the standard clinical imaging. Reconstructing TREIT images involves a solution of an inverse problem. The reconstruction is based on two steps: solving and updating an estimate of the dielectric property distribution through solution of an inverse problem. In this paper we consider a multi-GPU acceleration, which will allow us to significantly speed up the solution of the inverse problem. By conducting numerical experiments we compare results in assembling the Jacobian matrix by a CPU-based multiple computational cores server, a single GPU acceleration, and eventually a multi-GPU.