Research On Low-Dose CT Image Denoising Algorithm Based On Low-Rank Theory

Computed Tomography(CT)is a widely used imaging method in recent years.With the increasingly widespread use of CT scanning in various clinical fields,more attention has been paid to the radiation harm of CT scanning to the human body.Compared with other radiological examination methods,CT scanning brings much higher radiation than the others.Low dose scanning protocol technology can effectively reduce the amount of radiation received by the human body,and reducing tube current is currently the most commonly used method in clinical practice.However,low dose scanning protocols often cause degradation of the reconstructed image,leading to the formation of fringe artifacts.Artifacts usually have relatively prominent intensity characteristics that can significantly reduce the ability to distinguish between normal or pathological tissues,which can affect doctors’ clinical diagnosis.Therefore,how to obtain high-quality reconstructed CT images under low dose scanning protocols has important research significance.In order to improve the image quality of Low-dose Computed Tomography(LDCT),based on the noise characteristics of LDCT images,two image denoising algorithms are proposed in this paper based on the low-rank theory.The detailed research contents are as follows:(1)A two-stage denoising method for low dose CT images based on low rank matrix approximation is proposed.In this paper,the noise is layered according to its morphological characteristics.One layer is speckle noise without low rank property,and the other layer is fringe noise with low rank property.In the first stage,the weighted nuclear norm minimization(WNNM)method is used to extract image structure information,while removing speckle noise,to achieve the goal of removing only fringe noise from the weak noise in the image.Stripe noise has a contour and low rank is apparent.When rotating an image containing only stripe noise,the directionality of the stripe noise is concentrated in the vertical direction.At this time,the stripe noise has a lower rank than the image structure,which is referred to in the article as rotational low rank.In the second stage,the rotation low rank property of the fringe noise is utilized,and the low rank method is used again to achieve the separation of the fringe noise and the image structure to obtain the final denoised image.(2)A low rank and low dose CT image denoising method based on wavelet transform is proposed.When there are fringe noises in several directions in some local areas of the image,it is rough to deal with fringe noises directly by using the first algorithm proposed in this paper,which may lead to poor local denoising effect.Based on this,the Stationary Wavelet Transform(SWT)is introduced to decompose the image,realizing the fine processing of noise from different directions.Specifically,first,the WNNM algorithm is used to remove speckle noise,and then SWT is used to decompose the preliminary denoised image to obtain four component images: low-frequency,horizontal,vertical,and diagonal.Using the rotation low rank of fringe noise,the fringe noise removal of different directional components is carried out,and then the Inverse Stationary Wavelet Transform(ISWT)is used to obtain the final denoising image.