Application Of Residual Learning In CT Sparse Reconstruction Artifacts Correction

With the rapid development of science and technology,computed tomography(CT)equipment has been constantly upgrading,and has played an increasingly important role in the field of medicine and industry.In medical treatment,CT has the characteristics of high resolution and can provide clear images for all tissues and organs,so that medical personnel can quickly diagnose accurately.However,as a radioactive equipment,the X-rays of CT has a potential threat and has a great radiation hazard to the human body.In order to reduce the large dose of X-rays in CT scanning,there are three feasible methods: reducing the number of projections,reducing the scanning time and decreasing the voltage at both ends of the tube.Compared with the traditional method,this paper takes the reduction of projection number as a breakthrough point,and combines the current hot deep learning technology to study.In order to reduce the number of projection,the method of sparse reconstruction is mainly used in this paper.First of all,reconstruct from the original projection data by using the sparse reconstruction method,reducing the number of projection;secondly,establish the structure model about residual learning convolutional neural network to learn the artifacts feature from the spare reconstruction;finally,the CT image of the artifact-free is obtained by residual operation on the sparse reconstructed image,CT image artifact.In view of the characteristics of lots of X-rays and large radiation dose in traditional methods,a method based on sparse reconstruction combining with residual learning is proposed to correct artifacts.(1)Combining with the deep learning technology,using the residual learning convolution neural network to remove the artifacts in the sparse reconstruction.A residual learning convolution neural network model is established for training,learning and testing.According to the specific experimental data,it is verified that the residual learning convolution neural network is better than traditional methods,with shorter time and higher efficiency.(2)The residual learning convolution neural network model only uses a single convolution kernel mode.By changing the Goog Le Net model,we add it to the residual convolution neural network model structure,and transform the single convolution kernel mode into the multi convolution kernel mode.In this way,the improved Goog Le Net residual convolution neural network can better learn the feature of artifacts,and it will be better in the effect of removing artifacts.In the process of studying the artifacts in sparse reconstruction images,the experimental results show that this method greatly reduces the radiation dose of Xrays,effectively shortens the processing time and improves the effect of image reconstruction.