Research On Remote Bioimaging Technology Based On Deep Learning

In the field of medicine,how to apply advanced science and technology to human medical examination conveniently and cheaply,and improve human early prevention and treatment of diseases,is an urgent problem to be solved.Medical imaging technology is an important way to solve this problem.In electrical impedance tomography(EIT),body conductivity is obtained from voltage and current measurements on the boundary of the body.Compared with other imaging modalities,EIT is inexpensive,non-invasive and fast.EIT image reconstruction is ill-posed,nonlinear and under-determined.Traditional image reconstruction methods have limited ability to suppress noise interference and image artifacts.In this thesis,a remote biological imaging technology based on deep learning is proposed to improve it,which combines fast iterative algorithm with deep learning and realizes remote EIT imaging by telecommunication.The main works and innovations of this thesis are as follows:(1)For the initial reconstruction of EIT images,an improved iterative algorithm of sensitivity matrix(ISM)is proposed.Because the actual sensitivity matrix of electrical impedance imaging is obtained by uniformly distributed field,there is always a big deviation in the theoretical sensitivity matrix.This thesis provides an iterative algorithm to correct the sensitivity matrix.Firstly,a new global-difference model is constructed by using the sparsity of the perturbation of conductivity,so that the solution can jump out of the constraint of local optimal.Secondly,an iterative algorithm is designed to make the sensitivity matrix approach the theoretical value continuously.Finally,the normalization and hard threshold function are combined to eliminate noise and calculation error.The experimental results show that the proposed algorithm can reduce image artifacts and the influence of noise interference.(2)For the follow-up optimization of EIT initial reconstruction,an improved O-Net neural network structure is proposed.The traditional methods usually have problems such as high computational cost,loud noise and large modeling error.This thesis designs a deep neural network called O-Net to improve.Firstly,the ISM algorithm is used for initial reconstruction.The reconstructed image is used as the input of the neural network,so as to reduce the computational cost of the neural network.Secondly,the structure of O-Net neural network is constructed,and a connection is added between the input and output of the neural network to improve the learning efficiency.Finally,the sparse representation is constructed before the connection to remove the noise information in the input image.The experimental results show that the proposed O-Net neural network effectively reduces the computational cost and improves the quality of reconstructed images.(3)For the telecommunication of EIT,an improved remote imaging system of EIT is proposed.Because it is difficult for EIT system to calculate the measured data on a large scale in complex environment,this thesis adds a telecommunication module on the traditional EIT system.Specifically,firstly,the boundary voltage of the body is measured by the hardware equipment of EIT and converted into a digital signal.Secondly,the voltage data is converted into Modbus data protocol format,and then the secondary development interface or virtual serial port software is used to send and store the data in the cloud server.Finally,the back-end server reads the cloud data for EIT image reconstruction,and returns the reconstruction result to the mobile terminal for display.The experimental results show that the the proposed remote imaging system of EIT can effectively realize the telecommunication function and has good imaging effect.