Oracle-Net for nonlinear compressed sensing in Electrical Impedance Tomography reconstruction problems

Damiana Lazzaro,Serena Morigi,Luca Ratti

2024-04-09

Abstract:Sparse recovery principles play an important role in solving many nonlinear ill-posed inverse problems. We investigate a variational framework with support Oracle for compressed sensing sparse reconstructions, where the available measurements are nonlinear and possibly corrupted by noise. A graph neural network, named Oracle-Net, is proposed to predict the support from the nonlinear measurements and is integrated into a regularized recovery model to enforce sparsity. The derived nonsmooth optimization problem is then efficiently solved through a constrained proximal gradient method. Error bounds on the approximate solution of the proposed Oracle-based optimization are provided in the context of the ill-posed Electrical Impedance Tomography problem. Numerical solutions of the EIT nonlinear inverse reconstruction problem confirm the potential of the proposed method which improves the reconstruction quality from undersampled measurements, under sparsity assumptions.

Numerical Analysis

What problem does this paper attempt to address?

This paper mainly discusses the application of nonlinear compressed sensing in the reconstruction problem of Electrical Impedance Tomography (EIT). EIT is a non-invasive imaging technique used to reconstruct the conductivity distribution inside a medium from the voltage measured on surface electrodes. In the standard compressed sensing theory, linear measurements and sparsity assumption are crucial, but many practical problems involve nonlinearity.

The paper proposes a graph neural network (GNN) method called Oracle-Net to predict the support information of nonlinear measurements for promoting sparse reconstruction. Oracle-Net utilizes supervised learning to estimate the support of the exact solution from the measurement data. Then, the predicted support information is integrated into the regularization recovery model to solve the nonlinear inverse problem in the form of a constrained optimization problem. The paper provides error bounds for the approximate solution of the Oracle optimization problem and discusses these bounds in the context of the nonlinear inverse problem in electrical impedance imaging.

The paper also investigates the regularization problem of the nonlinear measurement model using the Proximal Gradient method and presents theoretical results demonstrating the convergence of the minimum of the regularization function towards the solution of the inverse problem as the noise level decreases. Additionally, the paper demonstrates the numerical solution of the proposed Oracle-based Proximal Gradient method for the EIT reconstruction problem, showing that the method can improve the reconstruction quality while reducing the number of measurements.

In conclusion, this paper aims to improve the reconstruction effect of electrical impedance imaging, especially in the case of insufficient data, by introducing Oracle-Net to solve the support estimation problem in nonlinear compressed sensing.