基于正弦图融合的CT金属伪影校正算法研究

Correction Algorithm for CT Metal Artifacts Based on Sinogram Fusion

目的为解决修复后的投影数据与周围投影数据之间过渡不连续的问题,提出一种基于正弦图融合的CT金属伪影校正算法。方法通过预处理和K均值聚类技术将具有相同空间信息的组织聚在一起生成先验图像,并根据金属区域与先验图像的投影差异校正原始图像投影以得到校正后的投影数据,最后采用滤波反投影算法重建得到校正后的CT图像。结果在CT仿真数据验证中,基于先验插值的金属伪影校正(Fusion Prior-Based Metal Artifact Reduction,FP-MAR)算法在单金属校正和多金属校正中的峰值信噪比分别为0.943和0.915,比线性插值校正金属伪影(Linear Interpolation Based Metal Artifact Reduction,LI-MAR)算法分别增加了28.65%和44.55%;FP-MAR算法在单金属校正和多金属校正中的结构相似性分别为0.984和0.961,比LI-MAR算法分别增加了48.41%和64.27%。临床CT伪影影像验证中,FP-MAR算法校正后CT金属伪影的主观评价高于LI-MAR算法校正后的CT金属伪影图像,且二者差异有统计学意义。结论本研究提出的算法可有效解决修复后的投影数据与周围投影数据之间过渡不连续的问题,更好地保留金属结构附近的信息。

Objective To propose a metal artifact correction algorithm based on sinusoidal graph fusion to solve the problem of transition discontinuity between repaired projection data and surrounding projection data.Methods By preprocessing and K-means clustering technology,organizations with the same spatial information were gathered together to generate prior images,and the corrected projection data were obtained by correcting the original image projections based on the projection differences between the metal regions and the a priori images.Finally,the filtered backprojection algorithm was used to reconstruct the corrected CT image.Results In the verification of CT simulation data,the peak signal-to-noise ratio of fusion prior-based metal artifact reduction(FP-MAR)algorithm based on prior interpolation in single-metal correction and multi-metal correction were 0.943 and 0.915,respectively.Compared with linear interpolation based MAR(LI-MAR)algorithm,it increased 28.65%and 44.55%,respectively.The structural similarity of FP-MAR algorithm in single-metal correction and multi-metal correction were 0.984 and 0.961,which were 48.41%and 64.27%higher than that of LI-MAR algorithm.In clinical CT artifact image validation,the subjective evaluation of FP-MAR-corrected CT metal artifacts was higher than that of LI-MAR-corrected CT metal artifact images,and the difference was statistically significant.Conclusion The algorithm proposed in this study can effectively solve the transition discontinuity between the repaired projection data and the surrounding projection data,and better preserve the information near the metal structure.