The Research On Low-dose CT Image Post-processing Algorithm With Improved Non-local Means

In recent years,X-ray computed tomography(CT)technology has developed rapidly and has been widely used in modern medicine.It has become an indispensable detection method in clinical medical diagnosis.However,in order to get high-quality imaging results in CT imaging,more radiation doses will often be added,which will cause injury to the health of the subjects.So low dose CT(LDCT)technology arises at the historic moment.It reduces the damage to the subjects by reducing the dose of CT radiation.But its drawback is that it is easy to reduce the imaging quality of reconstructed images.Therefore,imaging technology which can not only reduce CT radiation dose,but also obtain high-quality images with less noise has become a research hotspot in recent years.In view of the noise suppression of LDCT imaging,the post processing algorithm,which is not dependent on the projection data in the process of processing,can directly denoise the LDCT image reconstructed by the filter back projection algorithm,and become an important research direction to improve the quality of the LDCT image.This paper mainly focuses on the research of post-processing algorithms of LDCT images based on the recent Non-local Means(NLM)denoising algorithm.The main contents are as follows:1.The principle of Gauss filtering,bilateral filtering,anisotropic filtering and total variation denoising algorithm is expounded.The theoretical basis and advantages of the NLM denoising algorithm are introduced.Then the LDCT image is processed with the mentioned above five noise reduction algorithms,and the simulation experiment of LDCT brain model is carried out,and the advantages and disadvantages of the five algorithms are analyzed through subjective evaluation,objective evaluation and cross-sectional graph of the image after noise reduction.Finally,the advantages of NLM algorithm in improving the image quality of LDCT are verified compared with the traditional noise reduction algorithm.2.When the original NLM algorithm suppresses the noise of the LDCT image,the difference between the gray value of the pixel neighborhood is ignored,which leads to the blurring of the edge of the image and the loss of the details.Therefore,an improved NLM algorithm based on the gradient direction is given by using the similarity of the gradient direction between the two pixels neighborhood before and after the noise is added.Firstly,the denoised projection data is reconstructed by filtering back projection directly.Secondly,the reconstructed LDCT image is preprocessed by Gauss filter.Finally,the weights are constructed by using the gradient information and the mean value of the region,so that the new weights can better judge the similarity between the neighborhood blocks.Simulation experiments on the LDCT image of the pelvic bone and the actual LDCT image of the pelvic bone show that the improved algorithm can retain more details of the image and get better noise reduction performance.3.According to the strong directionality of strip artifacts in LDCT reconstructed image,a NLM algorithm based on direction matching is presented.The algorithm improves the original Gauss weighted Euclidean distance to the ellipse weighted Euclidean distance with anisotropic characteristics,and improves the original square search window into an elliptical search window.So as to improve the original NLM algorithm for image edge and smooth area pixel matching equal weight.The simulated brain model and the actual chest model are simulated.The experimental data show that the improved algorithm has a good balance effect in keeping the edge and suppressing the artifact.