Low Dose CT Imaging Studies Based On Sparse Representation

Due to its painlessness,high density resolution and accurate location of the lesions,X-ray computed tomography(CT)technique provides a reliable imaging data for clinical diagnosis and has been widely applied in clinical diagnoses.However,X-ray is ionizing radiation,which can cause physical damage when the dose exceeds the safe range,and has a certain probability of causing cancer and hereditary diseases.Owing to the radiation risk,interest has increased in the radiation exposure delivered to patients during CT examinations,and the issue of low-dose CT is becoming a research focus in image processing field.Reducing the radiation dose of CT scan can reduce the damage of X-ray to human body.However,the quality of CT reconstructed image is closely related to the radiation dose.If the X-ray dose is reduced,the imaging quality of the CT image will be significantly decreased.Therefore,it is of great academic significance and clinical application value to put forward effective low-dose CT imaging methods.There are three ways to reduce the radiation dose: limiting the scanning angle,reducing the intensity of the tube current and using the above two methods at the same time.In order to reconstruct an image with higher quality on the premise of reducing the dose of X-ray,three methods based on sparse representation theory are studied in this paper.In view of the above three cases,the imaging algorithms based on sparse representation theory are studied,and the quality of CT image is improved by using the algorithms.The main work is summarized as follows:1.Limited projection view will result in that the quantity of projection data is far less from number of pixels of the CT image to be reconstructed.In light of this situation,iterative reconstruction algorithms based on sparse representation regularization are proposed.The off-line dictionary trained by the clear CT images of different parts can capture contours,textures,edges and details of CT images effectively.The shearlets have the advantages of translation invariance and directional selectivity,so it can give a better representation of the detail information of the image.Sparse representation of intermediate images based on offline dictionary and discrete shearlet is added into the objective function as a regularization term by means of Augmented Lagrangian method so as to narrow down solution space and improve quality of reconstructed images.Through the Shepp-Logan simulation experiments and human CT experiments,it is verified that the two methods can still reconstruct the ideal image when the scan angle is less than 10% of the current commercial CT devices.2.Reducing the current intensity of X-ray tube will reduce the signal to noise ratio.For the low X-ray tube current intensity scanned reconstruction,penalized weighted least-squares approaches based on dictionary learning and discrete shearlet ware introduced to improve the performance of resisting noise in sinogram.Images can be sparse represented while the noise cannot.Most of the noise can be separated from the original image using the sparse transform.Therefore,the original image can be separated from the noise by using the sparse transform.The sparsity of the projection data is used as a penalty term to improve the stability of the solution.And in order to decrease the influence of noise statistical weight is added in objective function.The Shepp-Logan simulation phantom and the real human CT image verified denoising effect,and the proposed methods achieved satisfactory results.3.For the limited view and low X-ray tube current intensity scanned reconstruction,two statistical iterative reconstruction algorithms for low dose CT based on sparse representation regularization are proposed.Sparse representation of intermediate images based on off-line dictionary or discrete shearlet is added into the objective function as a regularization term so as to narrow down solution space and improve quality of reconstructed images.In addition,statistical weight is added in objective function to decrease the influence of noise on the effect of reconstruction.Compared with the above mentioned methods,the dose of X-ray was further reduced.The results of Shepp-Logan simulations and human CT experiments show that the two methods can achieve good imaging results when the projection data is very incomplete and the signal to noise ratio is very low.