Research On Regularization Method Of CT Image Reconstruction

Computed tomography?CT?is widely used in biomedicine,industrial nondestructive testing and other fields.After CT system acquires complete projections by scanning the object under examination?usually rotates continuously 360°?,filtered backprojection?FBP?algorithm can reconstruct high-quality CT images,which clearly show structures of the object.In practical applications,given the imaging geometry of CT system,CT canning scheme is often adjusted to reduce X-ray radiation dose,reduce scanning time and meet limitations of scanning conditions,common CT scanning schemes are few-view,limited-angle and low-intensity CT scanning.In order to reconstruct high-quality CT images,our main research work is to use appropriate prior knowledge to constrain the reconstructed images according to characteristics of CT reconstruction problems.Firstly,we propose a rotation center calibration method for CT system based on image gradient L₀ norm minimization to reduce geometric artifacts in reconstructed images.Geometric calibration of CT imaging system is one of the important steps in CT reconstruction,which aims to make projections consistent with actual imaging geometry and reduce geometric artifacts in reconstructed images.Micro-CT system is often used for 3D microscopic imaging of objects with small volume,which requires high-resolution CT images,so even a small transversal shift of the rotation center can also cause obvious geometric artifacts in the reconstructed images;these artifacts often appear as blurred or false edges,thus making the reconstructed image gradient L₀ norm larger.In this paper,the transversal shift of the rotation center is determined by minimizing the reconstructed image gradient L₀ norm on the basis of sinogram calibration.The validity of the proposed calibration method is verified by experiments on real micro-CT projections.Secondly,according to characteristics of multiple limited-angle?MLA?sampling?projection views are distributed in several segments in complete rotation angle?and distribution characteristics of the artifacts in reconstructed image,we propose multi-direction total variation minimization?MDTVM?method to reduce artifacts in the reconstructed image in MLA CT.MLA sampling can accelerate scanning speed and reduce radiation dose.In addition,the correlation between X-ray paths in MLA sampling is lower than that in general limited-angle sampling?projection views are concentrated in a limited rotation angle,usually less than 180 degrees plus fan angle?,and so it is easier to realize than few-view sampling?projection views are uniformly distributed in complete rotation angle?without frequently switching on and off the pre-collimator.However,in MLA CT,the images reconstructed by FBP suffer from some artifacts like shadow in several directions,and which are called the shading artifacts in this paper.We propose MDTVM method to suppress these artifacts.Compared with standard total variation minimization?TVM?,MDTVM enhances CT image sparsity in multiple directions,and is conducive to suppressing the shading artifacts.Thirdly,we introduce relative total variation?RTV?into CT reconstruction as a regularization term and propose RTV-regularized low-intensity CT reconstruction model,in addition,we use L-curve method to adaptively determine regularization parameter of the reconstructon model to improve reconstructed image quality.Low-intensity CT scanning can reduce radiation dose and scanning time.For example,in medical CT,low-intensity CT scanning can be achieved by reducing the X-ray tube current;in micro-CT,low-intensity CT scanning can be achieved by reducing X-ray exposure time.The term"low-intensity"means that the X-ray intensity is low and the projections contain high-level noise,in this case,images reconstructed by FBP suffer from obvious noise.To this end,we introduce the RTV into CT reconstruction model as a regularization term and propose the reconstruction method of RTV-regularized projection on convex sets?POCS-RTV?.RTV is defined on the basis of windowed inherent variation?WIV?and windowed total variation?WTV?.WIV can effectively depict the differences between noise and image structures in gradient domain;in image reconstruction,WIV value as the weight of WTV can adaptively penalize gradients corresponding to noise and image structures to suppress noise and preserve image structures.In addition,L-curve method is used to adaptively determine the regularization parameter of the reconstruction model,reducing the subjectivity of artificially selecting a regularization parameter,so as to acquire adaptive POCS-RTV?POCS-ARTV?.Experimental results verify the effectiveness of POCS-ARTV.Finally,we propose a limited-angle CT reconstruction model regularized by prior image induced relative total variation?piiRTV?to improve reconstructed image quality.In C-arm CT,dental CT,breast imaging and nondestructive testing of in-service pipelines in industry and so on,the CT system can usually only collect projections within a limited rotation angle due to the limitations of the scanning environment or the structures of the detected object.In this case,the CT images reconstructed by FBP suffer from serious shading artifacts.Prior image constrained compressed sensing?PICCS?method can reduce the shading artifacts in reconstructed images.When the signal-to-noise ratio of the prior image is high,PICCS can reconstruct high-quality CT images,otherwise,the CT image quality degrades.In order to inherit structure information of the prior image and reduce noise effects,we extract structure information of the prior image and apply it to limited-angle CT reconstruction to guide image structure recovery.The structure information of the prior image is determined by WIV,and on this basis we obtain the reconstruction model of piiRTV,which can effectively suppress noise and the shading artifacts.Experimental results verify the validity and superiority of piiRTV.