文章摘要

文章信息/Info

Title:: An electrical impedance tomography technique optimized by O-Net convolution neural network

作者:: 丁亚1; 王韦刚1△; 涂真珍2; 许晨东1; 1.南京邮电大学电子与光学工程学院微电子学院，南京 210023；2.南京邮电大学材料科学与工程学院，南京 210023

Author(s):: DING Ya1; WANG Weigang1; TU Zhenzhen2; XU Chendong1; 1.College of Electronic and Optical Engineering & College of Microelectronics, Nanjing University of Posts and Telecommunications, Nanjing 210023, China;2.College of Materials Science and Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210023

Keywords:: Medical imaging; Inverse problem; Finite element method; U-Net; Sparse coding; Random ellipse dataset

摘要:: 电阻抗成像逆问题的传统解决方法计算成本高、噪声大、建模误差大。本研究提出了一种O-Net深度卷积神经网络的反演方法，并与非迭代算法相结合以解决该问题。首先，通过截断奇异值分解方法对测量值进行图像重建，并将重建后的图像输入神经网络；其次，提出了O-Net神经网络结构，在输入与输出之间添加跳层连接，并进行1*1卷积，实现输入与输出的动态连接。最后，在跳层连接之前构建了稀疏表示方法，以减少输入图像中的计算误差。实验结果表明，相较于其它深度学习方法，本研究方法有效地减少了电阻抗成像的计算成本、提高了成像质量。同时实验结果验证了本研究的O-Net神经网络更适应于电阻抗成像技术。

Abstract:: Traditional solutions for the inverse problem of electrical impedance imaging suffers from high computational cost, large noise, and large modeling errors. We proposed a deep convolutional neural network of O-Net based inversion method and combined with non-iterative algorithm to solve this problem. Firstly, the measured values were reconstructed by truncated singular value decomposition method, and the reconstructed images were input into neural network. Furthermore, an O-Net network structure was proposed, in which a skip connection between the input and output was added and a 1*1 convolution was performed. Finally, a sparse representation method was explored before the skip connection, so as to reduce the calculation error in the input image. The experimental results showed that compared with other deep learning methods, this method effectively reduced the computational cost of electrical impedance tomography and improved the imaging quality. It also verifies that the proposed O-Net neural network is more suitable for electrical impedance imaging technology.