Thorax and internal organs boundary geometries determination using Convolutional Neural Networks in Electrical Impedance Tomography

Electrical Impedance Tomography (EIT) is a method used to produce images of the interior of a body by analyzing measurements of electrical potential taken along the boundary of the domain, obtained by applying current. This research focuses specifically on reconstructing the boundary of the domain and approximating the shapes of internal organs. The identified domain boundary and internal organ shapes can serve as inputs for traditional Gauss-Newton EIT inverse solvers, which require an initial estimation. To accomplish this, a Convolutional Neural Network (CNN) was proposed to predict geometric parameters using electrical potential measurements as input. This proposed approach is non-iterative, enabling rapid real-time outcomes. It is known that the contact conductivity in practical EIT typically varies. As a result, the experiments were structured in a way that the contact conductivity was randomly selected from a Gaussian distribution. An issue commonly encountered with standard EIT devices is the detachment of an electrode, which was simulated in this investigation. The results demonstrate that the proposed neural network surpasses baseline performance metrics, suggesting its ability to delineate the outlines of the thorax and internal organs to some extent, even in scenarios where an electrode measurement is absent.