Research On Intelligent Algorithm For Optimizing Spatial Resolution Of Electrical Impedance Tomograph

Electrical impedance tomography(EIT)is a new non-invasive imaging technology.Compared with other mainstream imaging technologies for lung disease diagnosis,EIT has the advantages of safety without radiation,fast response time and low cost.EIT has broad application prospects in the field of biomedicine.However,due to the "soft field" effect of EIT and the nonlinearity and morbid nature in the process of solving the inverse problem,the spatial resolution of the reconstructed images are low,and the reconstructed images are inaccurate,which limits the development of this technology,resulting in the EIT is still in the experimental research stage.In view of the above problems,based on the development prospect,research status,imaging characteristics and technical difficulties of electrical impedance tomography,this paper studies and analyzes the algorithms to improve the spatial resolution of electrical impedance tomography of the lung.The main work is as follows:1.To solve the problem of reconstruction image artifacts caused by the "soft field" effect and the problem of underdetermined in the visualization process of electrical impedance tomography.Combining the neighborhood information with the fast fuzzy C-means clustering algorithm,a new algorithm TR-NC for optimizing the artifact of lung electrical impedance tomography is proposed.And the algorithm is used to correct the artifacts in the reconstructed image of Tikhonov regularization algorithm.The simulations show that the TR-NC algorithm can effectively correct the artifacts of the reconstructed image by Tikhonov regularization algorithm and improve the quality of the reconstructed image.The correlation coefficient of the modified reconstructed image is increased by 24.75% on average as well as the relative error is reduced by 19.51%.Results of thoracic model experiments show that the TR-NC algorithm can detect the target in time and accurately when the target conductivity changes slightly.Compared with Tikhonov regularization algorithm,this algorithm has significantly improved the accuracy of the number,size and position of targets in the reconstructed images.2.Aiming at the nonlinear problem in the process of solving the EIT inverse problem.Combining the idea of depth learning with electrical impedance tomography,an EIT image reconstruction algorithm based on conditional Wasserstein generated adversarial network(CWGAN)is proposed.The algorithm combines the advantages of both conditional generative adversarial networks and Wasserstein generative adversarial networks to improve the stability of network training and the accuracy of generated images.The CWGAN network is trained by the constructed EIT data set to learn the nonlinear relationship between the boundary voltage value of the object field and the target conductivity distribution.The simulations show that the conductivity distribution reconstructed by this algorithm is closer to the real conductivity distribution,and the reconstructed image has fewer artifacts and clearer boundaries.Compared with the Tikhonov regularization algorithm,the correlation coefficient of the reconstructed image of the algorithm is increased by 72.56% on average as well as the relative error is reduced by56.25%.Compared with the TR-NC algorithm,the EIT algorithm based on CWGAN network improves the correlation coefficient of the reconstructed image by 32.34% on average,and the relative error is reduced by 21.42% on average.It proves that this algorithm can effectively improve the spatial resolution of EIT reconstructed images.3.Aiming at improving the low spatial resolution of imaging results of traditional EIT algorithm.Combining the traditional EIT algorithm with the CWGAN network idea,a new algorithm for optimizing the target of reconstructed images is proposed to further optimize the reconstructed image of the traditional EIT algorithm.The simulations show that the optimized conductivity distribution image of the target is closer to the real image,the image correlation coefficient is larger,and the relative error is smaller.It demonstrates that the CWGAN network structure can also be used to further optimize the reconstructed images of traditional EIT algorithm.