Effect of Measurement Noise on Reconstruction using Machine Learning with Electrical Tomography in the Case of the Abdominal Cavity

In this paper, we compare the reconstruction efficiency obtained by the least angle regression (LARS) and Elastic Net algorithms using electrical impedance tomography (EIT). Furthermore, we investigate the impact of measurement noise on the quality of reconstruction obtained by the more efficient algorithm. We reveal the relationship between the quality of reconstruction and the magnitude of information loss in a data frame. This study was conducted on a dataset representing EIT measurements for a cross-section of the abdomen at the bladder level. The simulated dataset contains 10,000 different measurement examples for a different number of inclusions.

I. Introduction

Computer diagnostic methods, such as EIT, allow for non-invasive imaging that can be used in medicine, industry, and other fields [1], [2]. The EIT technique uses electric current to reveal an object's internal structure in the spatial distribution of conductivity, which is done by solving an inverse problem or making predictions using machine learning models [3]–[7]. In this method, sampling is done by applying a small current to the tested object and then measuring its impedance and the voltage on its surface. The measurement data collected in this way make it possible to visualize the conductivity inside the object as well as to detect disturbances or the appearance of anomalies. This technique exploits the electrical properties of materials, but it has a relatively low spatial image resolution. The disadvantage of this method, the difficulty in producing results with good resolution, is primarily due to the restricted number of measurements. More than this, the sensitivity of the voltage measurement equipment is insufficient, yielding low sensitivity to variations of conductivity in certain areas.