基于深度学习的ClearInfinity算法对腹部体模CT扫描图像质量和辐射剂量的影响

Impact of ClearInfinity Algorithm Based on Deep Learning on Image Quality and Radiation Dose of Abdominal CT in Phantom

目的分析基于深度学习的ClearInfinity(CI)算法在上腹部体模CT扫描中降噪和提高图像质量的可行性,探讨优化图像质量的最佳权重比例。方法以同一管电压(120 kV)、不同容积CT剂量指数(Volume CT Dose Index,CTDIvol)水平(15、10、7.5、5、2.5 mGy)对腹部仿真体模进行扫描,每组图像均采用滤波反投影算法(Filter Back Projection,FBP)、Clear View迭代算法(CV算法,分别采用50%CV、90%CV重建)和深度学习算法(CI算法,分别采用10%CI、30%CI、50%CI、70%CI、90%CI重建)共8种不同方法进行重建,共获得40组图像。测量并记录各组图像肝脏、左侧肾脏及肌肉平均CT值、噪声和主观评分,同时记录各重建算法的重建时间。计算并比较CV算法、CI算法相比FBP算法降低噪声的能力。各组图像主观评分采用Kruskal-Wallis检验,重建时间、CT值和噪声比较采用单因素方差分析和Bonferroni校正进行多重比较。结果各组中不同部位CT值随不同算法权重比例的增高呈基本稳定状态(P>0.05);同一辐射剂量水平下,与FBP相比,随着CV及CI算法权重的升高,噪声水平显著下降(P<0.001);相同权重重建算法下,CI算法降低噪声的能力高于CV算法;主观评分比较,50%CV和10%CI、30%CI、50%CI算法图像质量能满足诊断要求。超低剂量水平(CTDIVOL=2.5 mGy)下30%CI图像质量最佳,且50%CI图像质量优于FBP重建(P<0.001)。各组中各重建算法的重建时间比较差异无统计学意义(P>0.05)。结论基于深度学习的CI算法较迭代算法显著降低了图像噪声,提高了图像质量,且30%~50%权重比例可以显著提高低剂量模式获得的图像质量。

Objective To evaluate the influence of ClearInfinity(CI)algorithm based on deep learning on image quality and radiation dose reduction of abdominal CT in phantom,and to investigate the optimal CI level.Methods Abdominal anthropomorphic phantom was scanned at different radiation dose levels of volume CT dose index(CTDIvol),which were set as 15,10,7.5,5 and 2.5 mGy,respectively.Images were reconstructed by filtered back projection(FBP),ClearView(CV)iterative reconstruction(CV algorithm,50%CV,90%CV)and deep learning algorithm(CI algorithm,10%CI,30%CI,50%CI,70%CI,90%CI)and 40 sets of images were obtained.The CT values,image noise of the liver and left kidney and subjective scores were measured,at the same time,we recorded the reconstruction times of all reconstruction algorithms.CV and CI algorithms were compared with FBP for the effect to reduce noise.Subjective scores of images in each group were compared using Kruskal-Wallis test,the reconstruction time,CT value and noise were compared with using one-way repeated-measures analysis of variance and the Bonferroni correction for multiple comparisons.Results The CT values of different parts in each group were basically stable with the increase of the weight ratio of different algorithms(P>0.05).At the same radiation dose level,compared with FBP,the noise level decreased significantly with the increase of the weight of CV and CI algorithm(P<0.001).Under the same weight reconstruction algorithm,the noise reduction ability of CI algorithm was higher than that of CV algorithm.Compared with subjective scores,the image quality of 50%CV,10%CI,30%CI and 50%CI algorithm could meet the diagnostic requirements.The image quality of 30%CI was the best at low dose level(CTDIvol=2.5 mGy),and the image quality of 50%CI was better than that of FBP reconstruction(P<0.001).There were no statistically significant dif ferences compared with the reconstruction time of all reconstruction algorithms(P>0.05).Conclusion Compared with CV algorithm,CI algorithm can provide a significant reduction in image noise and increase the image quality,and 30%to 50%CI can significantly improve the image quality under low dose mode.