Research On Low-dose CT Reconstruction Algorithm Based On Weighted Encoding With Sparse Nonlocal Regularization

CT technology plays a crucial role in detecting biomedical information.CT examination can deliver a true grayscale image of the cross-section and quantitative analysis by computer,which can help doctors enhance the accuracy of diagnosing patients.CT technology is a kind of high-energy X-ray imaging technology.X-ray radiation can induce harm to the human body,so it is important to reduce the dose of radiation.Low-dose CT technology can use lower doses of radiation to capture medical images,but lower doses inevitably lead to more noise in the image,affecting image quality and readability.Traditional two-stage image reconstruction algorithms find it challenging to completely separate Gaussian noise from impulsive noise,and often lead to unclear image or loss of detailed information.Therefore,how to reconstruct the image effectively while reducing the radiation dose has become a topic in the field of biomedical information detection and processing.Aiming at the reconstruction of low-dose CT images,this thesis establishes an image reconstruction algorithm with theoretical advantages and improves and optimizes it.The specific results are as follows: The specific results are as follows:(1)The weighted encoding with sparse nonlocal regularization algorithm,which can perform two image reconstruction tasks in a single step,has been introduced into the reconstruction of low-dose CT images to avoid the problems of reducing image clarity or losing detailed information due to the difficulty of completely isolating noise in traditional algorithms.The algorithm does not need a clear pulse pixel detection step and can suppress both Gaussian and pulse noise in mixed noise.(2)Combining sparse representation with low-rank constraint,a low-dose CT image reconstruction algorithm based on weighted encoding with sparse nonlocal regularization and low-rank constraint is proposed to solve the problem of noise removal when weighted encoding with sparse nonlocal regularization algorithm is applied to low-dose CT image reconstruction in Chapter 3.In the process of image reconstruction by weighted encoding with sparse nonlocal regularization,the problem of image reconstruction is solved by introducing proper regularization by using low-rank constraint as regularization.Experiments show that this algorithm has a better noise cancelation effect in the reconstruction of low-dose CT images.(3)Taking advantage of the advantages of edge-preserving median filter algorithm in preserving edge information,a new algorithm based on weighted encoding with sparse nonlocal regularization and edge-preserving median filter is proposed to solve the serious problem of missing edge information in low-dose CT image reconstruction in Chapter 4.Firstly,the image is reconstructed by using the edge-preserving median filtering algorithm,which keeps the edge information intact while removing some of the noise in the image.Then,mixed noise in the image is further processed by using weighted encoding with sparse nonlocal regularization algorithm.Experiments show that this algorithm has good ability of keeping edge information in the reconstruction of low-dose CT images.(4)Combining the advantages of the algorithms in Chapter 3 and Chapter 4,a combined reconstruction algorithm for low-dose CT images based on sparse nonlocal regularization weighted coding is proposed to further improve the denoising effect and the ability to retain edge information.Firstly,in order to improve the accuracy of noise recognition,the grey absolute correlation analysis algorithm is combined with the edge preserving median filtering algorithm to identify and mark edge points.Then,the algorithm and the reconstruction algorithm based on weighted encoding with sparse nonlocal regularization and low-rank constraint were combined to further improve the reconstruction effect.Finally,the greyscale of the edge points identified in the first step is assigned to the edge points in the reconstruction image,which further ensures the integrity of the edge information.For 10 different noise scenarios in the experiment,the proposed algorithm for low-dose CT images improved the peak signal to noise ratio by an average of about 0.9 d B and 1.7 d B compared to the previous two algorithms,and the average increase in structural similarity is about 38% and 62%.The reconstructed image is more similar to the original image as a whole,with clearer anatomical structures in the image and more complete information about the edges of tissues and organs.The proposed algorithm has advantages in noise removal and edge information retention,and can complete the image reconstruction well.In conclusion,the reconstruction of low-dose CT images is studied in this thesis,and several reconstruction algorithms of low-dose CT images are proposed.All proposed algorithms have the advantage of completing two image reconstruction tasks simultaneously within a single framework.Experimental results show that the proposed algorithm has advantages in noise removal and edge information retention,provides a viable method for reconstruction of low-dose CT images and further reduction of radiation effective dose for CT examination.