| An adaptive simultaneous algebraic reconstruction technique is proposed to settle the frequent and subjective adjustment of iteration step in conventional iterative reconstruction algorithms. Low dose CT imaging has gradually become a focus of international research in recent years. compressed sensing(CS) reconstruction algorithm is beneficial for does reduction of diagnostic by reducing the number of projections. Therefore, we propose an adaptive simultaneous algebraic reconstruction technique based on CS, which combine the advantage of SART and CS. Moreover, in order to reduce the beam hardening artifacts and improve the detection performance of soft tissue lesion, the dual energy decomposition algorithm is studied in this paper.In this paper, we first propose an adaptive simultaneous algebraic reconstruction technique(ASART) based on fuzzy entropy and simultaneous algebraic reconstruction technique(SART). After preliminary SART reconstruction, by quoting fuzzy entropy for edge detection of the reconstructed image which is used as prior information, a monotonous increasing function that defines the relaxation factor is constructed based on the neighborhood homogeneous measurement(NHM). Therefore, the proposed approach can select the relaxation fa ctor adaptively by the local character of the image. Then an ASART-TV algorithm is proposed by us which combine the very best of SART-TV method and ASART algorithm. The finite difference image of reconstructed image can be mapped to iteration step which represent the local character of the reconstructed image by the mapping function L what constructed in ASART-TV method. So, adaptive iterative process can be realized through substituting the iteration step for the adaptive step of ASART algorithm. In order to demonstrate the performance of ASART and ASART-TV algorithm, the projection data of Shepp-Logan phantom, anthropomorphic head phantom and patient are selected to conduct CT reconstruction. Normalized root mean square error(NRMSE), normalized mean absolute error(NMAE) and signal to noise ratio(SNR) are used for quantitative analysis of the reconstructed image of Shepp-Logan phantom, and Contrast(Con), standard deviation(STD) and contrast to noise ratio(CNR) are choose to estimate the reconstruction image of a nthropomorphic head phantom and patient quantitatively. Next, three dual energy decomposition algorithms which adopted different decomposition model and belong to different decomposition domain are selected to decompose the phantom which is built by us. Moreover, e lectron density error(EDR) is choose as a quantitative evaluation standard of decomposition image. At last, on the premise of making comprehensive consideration for the EDR, decomposition time and data requirements, this paper selects the iteration decomposition algorithm as our research focus. The decomposition image of line bar phantom and a nthropomorphic head phantom are selected to demostrate the superior performance of iteration decomposition algorithm.Experimental results show that the NRMSE, NMAE of ASART method decreased to 27.5%, 25.2% respectively, and the SN R of ASART method increase to 1.56 times of the SART algorithm on Shepp- Logan phantom. The STD of ASART-TV algorithm decreased to 83.7% and the CNR of ASART-TV increased to 1.14 times of the SART- TV method on clinical research. Besides, the ASART algorithm settle the problem of convergence fluctuation in SART method. However, the algorithm needs to be improved, because the selection of the mapping founction L is too crude. Finally, some optimum proposals are proposed by the analysis of reconstructed results. The EDR of the ROI1, ROI2 and ROI3 in the decomposition image using iteration decomposition algorithm are 3.25%, 0.19% and 0.02% respectively. Finally, least square estimation with smoothness regularization is used as noise constraint item and and this method shows superior performance on noise suppression with high image spatial resolution and low-contrast detectability. |
PDF Full Text Request