An Unsupervised Method for Evaluating Electrical Impedance Tomography Images

Abstract

Electrical impedance tomography (EIT) is an advanced visualizing technique with advantages such as real time, noninvasiveness, and low cost. However, the quality of EIT image still needs to be improved due to two inherent problems, i.e., ill-posed solution and “soft-field” effect. Nowadays, most of the existing EIT evaluation methods are supervised, relying on prior information or reference images, which are often unavailable in most applicable fields. In this paper, we proposed an unsupervised evaluation method based on a fast fuzzy C-means clustering algorithm. Compared with the most widely used evaluation methods, the new method could evaluate the reconstructed images without the need of any prior information or reference images. Simulations and experiments showed that the proposed method could efficiently evaluate the EIT images obtained using different algorithms with various parameters, and its capability of directly evaluating the reconstructed image made the evaluation process more useful in practical applications.