| It has been more than 50 years since the first CT came out in 1971.During these years,CT and its applications have developed rapidly with the progress of mathematics,engineering,and information technology.For example,the functions of equipment are from the initial CT for head scanning to devices mainly for general examination,from non-spiral CT to spiral ones primarily for vascular imaging,from single-row CT to multi-row systems mainly for cardiac coronary artery imaging,and from single energy CT to spectrum CT mainly for quantitative analysis of substances.These changes which contain scanning mode and imaging mode provide a lot of new grounds for clinical examination and accelerate the progress of clinical medicine.Since 2013,major manufacturers have accelerated the research and development of spectrum CT,such as the dual-source CT system of Siemens,the gem spectrum CT system of GE,and the dual detectors CT system of Philips.These products are obtained by monoenergetic CT through reforming the high-voltage generator,bulb,detector,data acquisition,and other components to achieve dual ray source dual detectors,kVp fast switching,single radiation source,dual detectors,kVp slow switching,and other functions.However,those spectrum CT implementation methods have the disadvantages of huge investment,long research and development cycle,complex system design,long patient scanning time,large radiation exposure,etc.Therefore,retaining the monoenergetic CT hardware,and adopting the method of reforming the data acquisition,software development,and operation mode is a cost-effective and lowinvestment spectrum CT implementation method.Using an iterative maximum likelihood polychromatic algorithm to reconstruct the spectrum of monoenergetic CT data is one of the effective methods.This method obtains the spectrum image under a specific keV by iterating on the conventional single CT scan data.Inspired by the sparse and low-rank image processing methods,the spectral reconstruction algorithm for multi-slice spiral CT is obtained by combining the monoenergetic CT scanning data collected by XHCT-16 multi-slice spiral CT system and the iterative maximum likelihood polychromatic algorithm.During the calculation,spectrum information is obtained by assuming two unknown quantities,called the photoelectric effect coefficient,and the Compton scattering coefficient on each point of the object.Therefore,if an object is discretized into J reconstruction points,there will be a total of 2J unknowns to be reconstructed in this algorithm,which is twice the unknowns of monoenergetic CT reconstruction.This will lead to a very large amount of calculation and a very long working out process.As a result,during each step of iterative reconstruction,we need to perform a sparse and low-rank operation on the data according to its characteristics so that it can be effectively reduced and optimized,then we carry out the next step of iterative reconstruction on the optimized data.Finally,we can get the spectral images of multiple slices at the same time.This thesis focuses on the research and discussion of spectral reconstruction combining iterative maximum likelihood polychromatic algorithm for multi-slice spiral CT using conventional scanning data.The main work is as follows:(Ⅰ)Accurate spectrum information is obtained by estimating the Bowtie filter,and the spectrum reconstruction algorithm of multi-slice spiral CT is accelerated based on CUDA architecture.The distribution of the XHCT-16 raw Bowtie filter in all directions is estimated by using blank scanning data,and accurate incident spectrum information before an X-ray passing through the object is obtained.To increase the operation speed of iterative reconstruction,CUDA computing architecture is used to optimize the spectrum reconstruction process of multi-slice spiral CT.Experimental results show that the reconstruction algorithm has greatly improved the accuracy of the spectrum and calculation speed through the processing of the above two processes,which provides favorable conditions for the development of clinical practice.(Ⅱ)A framelet-based spectral reconstruction algorithm for multi-slice spiral CT is proposed.Aiming at solving the problem of a large amount of data during the operation process,the photoelectric effect coefficient and Compton scattering coefficient of each slice in the iteration process are transformed into the wavelet domain for sparse representation,then the data are sparse and the noise generated by the data in the iteration process is reduced by using the threshold processing method.Finally,the transient data are transformed into an image field by inverse framelet transform to obtain the processed photoelectric effect coefficient and Compton scattering coefficient and then carried to the next iterative reconstruction.The experimental results show that,compared with the reconstructed images of monoenergetic multi-slice spiral CT,our method can integrate the single scan data with the same conventional CT,and reconstruct the photoelectric effect images,Compton scattering images,monoenergetic images synthesized according to different keV values and color spectrum images of multiple slices at the same time,and our reconstructed images are better than the reconstructed images of conventional CT both in visual observation and objective measurement.(Ⅲ)A spectral reconstruction algorithm for multi-slice spiral CT based on 3D shearlet is proposed.According to the sparsity between adjacent slices in the reconstruction process of multi-slice spiral CT,a three-dimensional shearlet transform is used to transform the slices in the iterative process as photoelectric effect coefficient and Compton scattering coefficient into the frequency domain.After threshold processing,the transient data are inversely transformed into the image field to obtain the processed photoelectric effect coefficient and Compton scattering coefficient,then they are brought into the next iterative reconstruction.The experimental results show that the photoelectric effect image,Compton scattering image,single energy image synthesized according to different keV values and color spectrum image reconstructed by our method are better than conventional CT reconstruction images and compared reconstruction methods in terms of visual observation and objective measurement.(IV)A spectral reconstruction algorithm for multi-slice spiral CT based on tensor-based dictionary learning is proposed.According to the characteristics that the attenuation coefficient of an object in spectrum CT can be expressed as the linear representation of several spectrum components and the correlation between adjacent slice images,the photoelectric effect coefficient and Compton scattering coefficient in the iterative reconstruction process are rearranged by using the spectrum information and geometric information at the same time,and the rearranged data are regarded as a tensor.Then,the tensor dictionary learning algorithm is used to process the low-rank data,and the obtained data are expanded into photoelectric effect coefficient and Compton scattering coefficient,then carried into the next iterative reconstruction.The experimental results show that the photoelectric effect image,Compton scattering image,single energy image synthesized according to different keV values and color spectrum image reconstructed by our method are better than conventional CT reconstruction images and compared reconstruction methods in terms of visual observation and objective measurement. |
PDF Full Text Request