Fast 3-D Electrical Impedance Tomography Imaging of Tumor Boundary Based on Plane Extension Layer and Deep Learning

A 3-D electrical impedance tomography (3D-EIT) method is proposed for tumor boundary based on deep learning and plane extension layer (PEL). First, a network is developed for imaging that combines an encoder-decoder and a spatial pyramid pooling module with dilated convolution. Second, a PEL-based preprocessing method for voltage data is proposed to generate a larger 2-D voltage data matrix to achieve the same resolution as the network outputs while preserving the original information of the data. Third, the effect of ResNet backbone network layers on imaging accuracy and network model anti-noise ability is further explored, resulting in a fast and high-precision method for tumor boundary imaging. The performance of the proposed method is verified through simulations and experiments. The imaging algorithm proposed in this study achieves an image correlation coefficient of ICC = 0.8068 on the numerical simulation results and an image correlation coefficient of ICC = 0.836 on the experimental results. The minimum image reconstruction time is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${t} \; = 0.013$ </tex-math></inline-formula> s. In addition, the PEL method proposed in this study can compress the training weight file by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\delta \; = 1$ </tex-math></inline-formula> MB. The results show that the 3D-EIT method proposed in this study is able to rapidly and accurately present tumor contour boundaries and locations, thus promising to help surgeons achieve rapid intraoperative tumor margin detection and reduce the risk of postoperative recurrence.