Research On Iterative Reconstruction Algorithm Acceleration And Reconstruction Image Denoising Algorithm

Computed Tomography(CT)imaging technology uses the attenuation of X-rays through the object to obtain the internal information of the object,which has the advantages of non-destructive,intuitive and accurate,and is widely used in industrial inspection,medical imaging diagnosis and other fields.In practical applications,X-ray energy can cause radiation damage to humans or due to the large size and complex structure of industrial workpieces under inspection,only reduced dose or sparse angle projection data can often be obtained,however,because CT systems also introduce a lot of noise during the scanning process,the iterative reconstruction algorithm can have a good suppression effect on noise,iterative algorithms are better than parsing algorithms for reconstruction of less projected data,but their use in practice is limited by their large computational size.Because of the sparse angular projection data,the quality of the reconstruction may not be as good as we need it to be,the reconstructed image needs to be denoised.In addition,deep learning has achieved very good results in the field of computer vision,so some scholars have started to introduce deep learning into the field of medical imaging,and proposed to combine deep learning with CT image denoising.Therefore,this paper addresses the problems of computationally intensive iterative algorithms and the inclusion of noise in CT images,with the following main elements:1.The reconstruction principle and reconstruction steps of the ART reconstruction algorithm are introduced in detail,the main factors affecting the reconstruction quality and reconstruction speed of the iterative reconstruction algorithm are analyzed,and improvements are made to the calculation method of the weight factor,and the corresponding grid intersection length is proposed to be solved quickly through the corresponding proportional relationship,comparing the improved method of this paper with the reference method,the experimental results show that the reconstruction speed of the improved method is faster than that of the reference method.Finally,the improved method of calculating projection coefficients is extended to SART,ML-EM,OS-EM and OS-SART,with a speed improvement compared to the method in the literature [42].2.The most direct way to improve the quality of CT images is to denoise the images in the image domain.Classical denoising algorithms can improve the quality of CT images,but are prone to problems such as artifacts and step effects.As deep learning has made greater progress in the image domain,many neural network-based denoising models have been introduced to the CT denoising problem.Therefore,in this paper,the RED-CNN network model was used to complete the denoising of low-dose CT images,and then it was introduced into industrial CT images,and through the experimental results,it was found that a better denoising effect could also be achieved.