Information Enhancement With Multilayer Convolutional Neural Network for Accurate Lung Imaging

Abstract

Electrical impedance tomography (EIT) is a novel imaging technique for lung monitoring. Due to traumatic injuries or surgical reasons, the number of electrodes for current injection and voltage measurement may be limited causing inadequate data. Thus, the information related to conductivity distribution cannot be accurately deduced from the limited measured data. To obtain high-quality lung images when the number of electrodes is limited, a new information enhancement method is proposed. 2-D thorax models with eight electrodes and sixteen electrodes are built, respectively. The mapping between the voltage data of the two kinds of models is established. With this method, the voltage data measured by the eight-electrode lung EIT can be mapped into the equivalent voltage data of the 16-electrode lung EIT. The results show that the voltage data after information enhancement is almost the same with the target voltage data. In comparison to the reconstructed image with the eight-electrode data, image reconstruction shows a large improvement when using the enhanced data. The effectiveness of the proposed method is also testified in the presence of noise interruption and lung variation. It is found that the proposed method has strong immunity to noise and performs well when the lung shape varies.