Research On Spare Angle Cone-beam CT Reconstruction Algorithm

Computerized tomography(CT)is an advanced medical imaging technology that can detect the structure,shape,location,and precise size of objects inside in a non-contact and non-destructive manner,and is widely used in medical diagnosis and industrial nondestructive testing.Cone-beam CT uses cone-beam X-ray source to illuminate the object to be reconstructed,which can achieve the imaging effect of the object with isotropic resolution in three-dimensional space,and the acquisition speed is faster.However,the X-rays emitted by CT examinations are harmful to humans,and patients with high radiation doses have a significantly higher chance of causing cancer than average.Reducing the radiation dose as reasonably as possible without damaging the reconstruction quality is the key to studying CT reconstruction problems.In this paper,the problem of sparse-angle cone-beam CT reconstruction with increased sampling angle interval and reduced projection angle is studied in depth.The main research contents are as follows:(1)Aiming at the situation that there have staircase effect in the reconstruction results based on the total variation minimization reconstruction model,a cross-iterative cone-beam CT reconstruction framework based on half quadratic splitting is proposed.After replacing the variable of the regularization term in the sparse-angle cone-beam CT reconstruction model,the reconstruction optimization problem is converted into two easy-to-solve sub-problems through half quadratic splitting.One of the sub-problems mainly solves the data consistency constraint,which is equivalent to a well-defined linear optimization problem,and can be solved by matrix pseudo-inverse;the other sub-problem is a denoising problem,which can be solved using various Gaussian denoisers.Half quadratic splitting decouples the reconstruction problem and reduces the computational difficulty.The reconstruction results can be obtained by solving these two sub-problems in turn and iterating.(2)For the denoising problem in(1),this paper uses the non-local self-similarity of cone-beam CT,based on the principle of low-rank matrix recovery,uses a non-local clustering strategy to form an image similarity block matrix as the basic processing unit,and the original weighted nuclear norm minimization is extended to three-dimension,making it possible to implement cone-beam CT denoising.Since Schatten p-norm is a better expression of matrix sparsity,the denoising algorithm based on Schatten p-norm minimization is further extended and applied to cone-beam CT denoising.Experimental results show that compared with other algorithms,the algorithm that we proposed effectively overcomes the staircase effect and has achieved good results in visual inspection and quantitative evaluation.(3)A sparse-angle cone-beam CT reconstruction algorithm based on guided image filtering is proposed.Since the guided image filtering can transfer the characteristics of the guided image to the target image,based on the edge preservation of the SART algorithm and the suppression artifacts characteristic of the TpV minimization algorithm,this paper uses the SART reconstruction result as the filter input,and the TpV minimized reconstruction result as the initial guide image.Dynamically update the guide image to guide the image output,so that the reconstruction results have achieved good results in terms of edge preservation and artifacts reduction.