基于多层人工神经网络的电阻抗成像算法

Image reconstruction method based on multilayer artificial neural network for electrical impedance tomography

电阻抗成像(electrical impedance tomography,EIT)作为一种非侵入式的医学成像技术,其重建过程是一个难以计算的病态逆问题。为了保证EIT成像精度并提高运算速度,设计了基于多层神经网络(multilayer artificial neural network,MANN)的电阻抗成像逆问题求解方法。该方法分为两个步骤,首先利用EIT正问题得到MANN的训练数据,随后设计MANN网络,经过调参和训练后,该方法能迅速得到精度高的结果。该方法与牛顿法和分裂Bregman方法比较,对仿真数据和实测数据均得到良好的效果。

Electrical impedance tomography(EIT)is a non-invasive medical imaging technique.However,the reconstruction problem involves an ill-posed inverse problem,which is difficult to calculate.In order to ensure the accuracy of EIT imaging and increase the operation speed,this study explores the inverse problem-solving method of EIT based on multilayer artificial neural network(MANN)through machine learning.The methods of this studyare divided into two steps:Generate training data and design the MANN to obtain impedance distribution.Results were compared with the NewtonRaphson method(NRM)and split Bregman method(SBM).A series of experiments indicated that the proposed method outperforms the NRM and SBM for the simulation data and measured data.