Karl、DL重建算法对上腹部CT图像质量和辐射剂量的影响  

The effect of KARL and DL reconstruction algorithms on image quality and radiation dose of upper abdominal CT

作　　者：王诗瑜 刘义军[1] 李贝贝 赵明月 王旭[1] 张竞颐 WANG Shi-yu;LIU Yi-jun;LI Bei-bei(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Liaoning 116011,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁116011

出　　处：《放射学实践》2023年第5期626-630,共5页Radiologic Practice

摘　　要：目的:探讨不同级别Karl迭代重建技术和深度学习(DL)重建算法对上腹部CT图像质量及辐射剂量的影响。方法:应用联影uCT-760对腹部仿真模体(CDP-2)分别采用10~150 mAs(间隔20 mAs)行上腹部CT扫描,管电压120 kV,螺距0.9875,转速0.5 s/r,扫描层厚5 mm,层间距5 mm,重复扫描三次。对原始数据分别采用滤波反投影法(FBP)、Karl (1~9级)、DL(1~4级)算法重建1 mm图像。在轴面图像上测量肝脏及同层腹壁脂肪的CT值和SD值并计算信噪比(SNR)和对比信噪比(CNR),记录容积CT剂量指数(CTDIvol)和剂量长度乘积(DLP)。以常规剂量(150 mAs)FBP重建图像为对照,由两位观察者采用五分法评估各组重建图像的主观质量并行kappa一致性检验,采用Kruskal-Wallis H检验比较各组主客观图像质量。结果:各组FBP、Karl(1~9级)、DL(1~4级)重建图像的CT值差异无统计学意义(P>0.05)。随着Karl/DL等级的增加,相应组别的SD值降低(P<0.05),SNR和CNR升高(P<0.05)。两位观察者对各组图像主观评分的一致性较好(kappa=0.721~0.837),50 mAs(DL-2)、70 mAs(Karl-6)图像与150 mAs的FBP图像质量差异无统计学意义(P>0.05),辐射剂量分别降低了53.3%、66.7%(P<0.05)。结论:与常规剂量(150 mAs)FBP算法相比,在获得相同图像质量情况下,应用Karl和DL算法能显著降低辐射剂量。Objective:To explore the effects of Karl's(Keep Artifact Really Low)iterative reconstruction technique and deep learning(DL)reconstruction algorithm on upper abdominal CT image quality and radiation dose at different levels.Methods:The abdominal simulation phantom(CDP-2)was scanned with a UCT-760 scanner at 10~150mAs(20mAs interval)and 120kVp.The scanning parameters included pitch of 0.9875,rotational speed of 0.5s/r,slice thickness of 5mm and slice interval of 5mm.The scanning was repeated three times.FBP,Karl(level 1~9)and DL(level 1~4)algorithms were used to reconstruct images with 1mm thickness from the original data.CT and SD values of liver and abdominal wall fat were measured on axial images,and SNR and CNR were calculated.CTDIvol and DLP were recorded as well.The images reconstructed by FBP at the conventional dose(150mAs)were considered a control group.Two observers were selected to evaluate the subjective quality of reconstructed images in each group with the quintic method and Kappa consistency test.The Kruskal-Wallish test was conducted to compare the subjective and objective image quality.Results:The CT values of images reconstructed by FBP,Karl(grade 1~9)and DL(grade 1~4)demonstrated no difference among all groups(P>0.05).With the increase of Karl/DL grade,the SD value of corresponding groups decreased(P<0.05),whereas SNR and CNR increased(P<0.05).The subjective scores of the two observers in each group were consistent(Kappa=0.721~0.837).No significant statistical difference was found between the image quality of 50mAs(DL-2),70mAs(Karl-6)and 150mAs(FBP)(P>0.05),and compared with FBP,the radiation dose of DL-2 and Karl-6 decreased by 53.3%and 66.7%,respectively(P<0.05).Conclusion:Compared with conventional FBP(150mAs),Karl and DL algorithms can significantly reduce the required radiation dose while obtaining the same image quality.

关 键 词：腹部模体 Karl算法 DL算法 辐射剂量 图像质量 

分 类 号：R814.42[医药卫生—影像医学与核医学]