Accelerating magnetic induction tomography‐based imaging through heterogeneous parallel computing

Abstract

Magnetic Induction Tomography (MIT) is a non-invasive imaging technique, which has applications in both industrial and clinical settings. In essence, it is capable of reconstructing the electromagnetic parameters of an object from measurements made on its surface. With the exploitation of parallelism, it is possible to achieve high quality inexpensive MIT images for biomedical applications on clinically relevant time scales. In this paper we investigate the performance of different parallel implementations of the forward eddy current problem, which is the main computational component of the inverse problem through which measured voltages are converted into images. We show that a heterogeneous parallel method that exploits multiple CPUs and GPUs can provide a high level of parallel scaling, leading to considerably improved runtimes. We also show how multiple GPUs can be used in conjunction with deal.II, awidely-used open source finite element library. KEYWORDScomputational electromagnetics, magnetic induction tomography, parallel applications INTRODUCTIONMagnetic Induction Tomography (MIT) is a non-invasive imaging technique, which has applications in both industrial and clinical settings. In essence, it is capable of reconstructing the electromagnetic parameters of an object from measurements made on its surface. These parameters are the permittivity, , the permeability, , and the conductivity, . An MIT device consists of two sets of coils placed around the boundary of the object to be imaged. The first set of coils is used for the purpose of excitation, and by passing a current through each coil in turn, a primary magnetic field is created. The second set of coils is then used for measurement. This procedure causes an eddy current when each of the primary magnetic fields interacts with a conducting body inducing secondary magnetic fields, and hence voltages, that are measured in the second set of coils. In order to estimate the electromagnetic properties of the material, ( , , ), from the induced currents and measured voltage, an inverse problem must be solved. In many practical applications, the distribution of one or more of these material parameters is assumed to be constant throughout the medium of interest.Conventional imaging techniques for imaging cerebral stroke, such as Magnetic Resonance Imaging (MRI) and Computer Tomography (CT), are expensive. Although MRI may be used for real-time image reconstruction, 1 it has been proposed that MIT can offer a low cost alternative in the first stages of diagnosis. 2 However, the low conductivity contrast between biomedical tissues presents significant challenges to MIT, and there are considerable difficulties in employing current computational techniques to solve the associated inverse problem. 3,4 Enabling MIT to take the step from being an experimental technique, which has already received some clinical interest, to become a viable imaging technique for the detection and monitoring of conditions, such as cerebral stroke, requires a step change in the quality of ...

Application of a Generative Adversarial Network in Image Reconstruction of Magnetic Induction Tomography

“…where V is a vector of the measured phase of boundary voltage, F is the forward model map of the internal conductivity distribution, which is usually linearized to the sensitivity matrix in traditional algorithms as V = Sσ. The purpose of the MIT image reconstruction (an ill-posed, nonlinear inverse problem) is to obtain the conductivity distributions by solving the model in Equation (2). The inverse problem is generally expressed as σ = F −1 (V) or σ = S −1 V. Furthermore, the sensitivity matrix cannot accurately describe the corresponding relationship between the measured value and conductivity due to the soft field effect of electromagnetic field, which increases the difficulty of solving MIT problem.…”

“…Firstly, the generator is pre-trained, and then it is connected to the discriminator for GAN training. Having obtained the voltages V (1) , V (2) , . .…”

“…where α is learning rate. After hyperparameters selection finished, a needed crossvalidation set as (V (1) , σ (1) ), V (2) , σ (2) , . .…”

“…Magnetic induction tomography (MIT) is a kind of electromagnetic imaging technology based on the eddy current testing principle [ 1 ]. In general applications, many groups of excitation–detection combination coils are used to obtain induced magnetic field signals near the imaging body [ 2 ], and then the conductivity distribution of the range to be measured is calculated by the reconstruction algorithm [ 3 , 4 ]. Because of its non-invasive and non-contact characteristics, MIT is suitable for geological exploration [ 5 ], industrial flaw detection [ 6 ], impurity detection [ 7 ], and medical imaging [ 8 ].…”