Computerized Tomography(CT)can provide multi-directional anatomical images of the various parts of the human body without tissue overlapping in a non-invasive way.And it is one of the most widely used imaging methods in clinical practice.However,the X-ray radiation it produces during image acquisition may cause non-cancer diseases and cancer risk.With the attention of doctors and patients to radiation problems,low-dose CT scanning technology is gradually being applied to the clinic.In actual scanning,the radiation dose is often reduced by reducing the tube current.However,the projection data scanned by this method is strongly interfered by noise,resulting in a decrease in the quality of the reconstructed image,thereby affecting medical diagnosis and treatment.Therefore,research on low dose CT images denoising has important clinical application value.Based on simulation data,phantom data and clinical data,this paper studies the LDCT noise removal and artifact suppression methods from the projection domain and image reconstruction domain,aiming at improving the quality of low-dose CT reconstructed images.The main work is as follows.(1)For the High-intensity noise problem caused by reducing tube current,this paper studies the statistical iterative denoising method based on total variation minimization.Firstly,the projected data is preprocessed to obtain the initial intensity of the incident ray,and the projected data is calibrated according to the consistency coefficient of the detector.Then,according to the total variation minimization denoising model,statistically iterative denoising is performed on the projection data after preprocessing.The denoised projection data is reconstructed by Filtered Back Projection(FBP)to obtain a reconstructed image with a lower noise level.This study suppresses noise in the projection domain,avoiding its mapping to the entire image domain during reconstruction.We applied this algorithm to clinical head and lung projection data,which significantly improved the reconstructed image quality.(2)The denoising algorithm for the projection domain is highly dependent on the data acquisition format of the CT vendor,and the portability is poor.We explore the noise removal and artifact suppression methods in the image domain.The Weighted Nuclear Norm Minimization(WNNM)algorithm was combined with the Block-matching and 3D filtering(BM3D)algorithm to propose the BM3D-WNNM image post-processing method.The proposed method has a good removal effect on both speckle noise and strip artifacts.And it improves the removal performance of LDC noise and artifacts.(3)Aiming at the poor effect of traditional image post-processing denoisir method on the removal of strong artifacts and the shortcomings of artifact suppressic dictionary learning method,this paper proposes a differential noise artifact suppressic method based on online dictionary learning.The proposed method designs two mod for training different types of training samples,enhances the flexibility of trainir dictionary samples,and avoids the subjective influence of manual acquisition of noi samples.Secondly,the online dictionary learning algorithm suitable for large samp training is used for dictionary training,which improves the training speed.Finally,t training method proposed by the algorithm to obtain the noise dictionary by training t high and low dose phantom data can provide its unique noise calibration dictionary f different machines and improve the pertinence of the algorithm. |