Research On Intelligent Reconstruction Algorithm Of CT Image

As a high-performance imaging technology,computed tomography(CT)has been widely used in the field of disease diagnosis because it can obtain images of objects' internal structures without any destruction.However,many problems appear as the CT technology is used more and more widely.Studies have shown that every year there are many patients who get cancer just because of exposure to too much radiation.Therefore,it becomes a hot spot topic recently to reconstruct high quality CT images with lower radiation dose.As we all know,the quality of reconstructed CT image is highly related with CT dose.Usually,reconstructed images with lower radiation dose,based on traditional reconstruction algorithms,are unable to meet the clinical requirements.In addition,for some special applications,single energy CT cannot meet practical requirements and thus dual energy CT with stronger discriminative ability is utilized instead.In the field of safety inspection,dual energy CT is used to check drugs,explosives and other contraband.In the field of medical diagnosis,dual energy CT is used to measure bone mineral density and iron concentration in the liver.Although dual energy CT has been widely used in many fields,it still has some problems such as low accuracy and artifacts.Therefore,it is necessary to improve the reconstruction algorithms to achieve high quality dual-energy CT images.The main work of this thesis can be summarized as follows:(1)There are two kinds of low dose CT: low exposure CT and sparse view CT.Iterative reconstruction algorithm can be used to reconstruct high quality sparse view CT image,but its limitations of slow reconstruction speed and difficulty of parameter selection restrict its applications in clinical medicine.To solve this problem,we propose to generalize the iterative reconstruction steps of the alternating direction method of multipliers(ADMM)into the form of iterative reconstruction networks,which are used to reconstruct CT images.Three different ADMM networks are presented according to the degree of generalization.In the training stage,the networks learn the iteration parameters,regularization terms and other indefinite terms which are necessary in the iterative reconstruction step,thus greatly reducing the complexity of parameter adjustment and model construction.Finally,the performances of the proposed networks are evaluated based on different evaluation criteria.Experimental results show that the proposed networks can effectively reduce the streak artifacts in reconstructed images.Compared with the traditional algorithms,ADMM networks can reconstruct higher quality CT images.(2)Nowadays,the reconstruction networks used for dual energy CT are all based on a single domain,namely,either image domain or projection domain.Generally,the dual energy CT reconstruction algorithm based on projection domain has serious noise problem,and the dual energy CT reconstruction algorithm based on image domain can not remove beam hardening artifacts.Therefore,a joint domain dual energy CT image reconstruction network is developed,by integrating the image domain network with the projection domain network via a domain conversion module,which can achieve highquality dual energy CT image reconstruction by extracting the characteristics of the joint domain.Experimental results indicate that dual energy CT images reconstructed with the proposed joint domain reconstruction network has less artifacts and higher accuracy.Moreover,the proposed algorithm takes less reconstruction time.