Sparse-view CT Reconstruction Based On Improved Residual Network

Computed tomography(CT)combines a series of X-ray images taken from different angles of the human body with computer processing to create cross-sectional image slices of bones,blood vessels and soft tissues in the body.Compared with regular X-rays,CT scan images provide more detailed information.CT scans can be used to visualize almost all parts of the human body,and can be used to diagnose illness or injury and to scheme medical treatment,surgical,or radiation treatments.During a CT scan,the human body is briefly exposed to ionizing radiation.Because a CT scan collects more detailed information,the amount of radiation is greater than when exposed to regular X-rays.Although no related studies have shown that low-dose radiation used in CT scans can cause long-term harm,the potential cancer risk may increase slightly at higher doses.Therefore,we hope to get clear imaging while reducing radiation damage.This article has conducted in-depth research on the reconstruction of sparse angle CT images,the main work of this article is as follows:With the development of CT imaging technology,people have put forward higher requirements for the quality of CT image reconstruction.While meeting imaging quality requirements,it is desirable to use the lowest possible dose of radiation.Sparse view reconstruction is an effective measure to solve the problem of radiation dose.Because the angular range of the projection data does not meet the data integrity conditions,sparse view reconstruction has always been a difficult problem in CT image reconstruction.In this paper,we mainly propose a CT sparse view reconstruction algorithm based on residual network.We first improve the traditional residual network,making the program faster,while ensuring the quality of images,and to a certain extent,eliminating artifacts caused by sparse view projections.Second,when multiple residual modules are stacked in the network,it is easy to cause gradient explosion as the network deepens,our solution is to add a scale factor to the residual module to prevent this situation.From the experimental results,it can be seen that this method can ensure the removal of artifacts while retaining complete details and sharp edges.When the CT images are sparsely sampled,the data volume of the obtained projection images will be relatively small.One of the solutions is to use the sine diagram interpolation method to process the sparse sine diagram and generate denser sine diagrams,but this will affect the quality of the reconstructed slice.This paper proposes a new model,which is based on U-Net and residual learning sparse view CT sine graph interpolation method.The sine image is inserted into a filtered back projection(FBP)algorithm for reconstruction by neural network.U-Net can effectively learn the complex structure of sine diagram data,especially for biomedical images,residual learning can learn the difference between the input sine diagram data and the full view sine diagram data,insteadof directly transforming the sine diagram data.Compared with other model algorithms,the experimental results show that even when the sine graph data reaches 90% sparsity,good results can be achieved.