三种不同重建算法在低剂量胸部CT图像质量中的比较  

Comparison of image quality of low dose chest CT using three different reconstruction algorithms

作　　者：冯倩倩 于卫军[1] 唐威[1] 赵心明[1] 黄遥[1] 吴宁[1,2] FENG Qian-qian;YU Wei-jun;TANG Wei;ZHAO Xin-ming;HUANG Yao;WU Ning(Department of Diagnostic Imaging,National Cancer Center/National Clinical Re-search Center for Cancer/Cancer Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100021,China;PET-CT Center,National Cancer Center/National Clinical Re-search Center for Cancer/Cancer Hospital,Chinese Academy of Medical Sciences and Peking Union Medical College,Beijing 100021,China)

机构地区：[1]北京协和医学院/国家癌症中心/中国医学科学院肿瘤医院影像诊断科,北京100021 [2]北京协和医学院/国家癌症中心/中国医学科学院肿瘤医院PET-CT中心,北京100021

出　　处：《中国肿瘤临床与康复》2021年第12期1409-1413,共5页Chinese Journal of Clinical Oncology and Rehabilitation

基　　金：国家重点研发计划重大慢性非传染性疾病防控研究重点专项(2017YFC1308700);中国医学科学院医学与健康科技创新工程(2017-I2M-1-005);中国癌症基金会北京希望马拉松资金资助项目(LC2018A03)。

摘　　要：目的探讨传统滤波反投影法(FBP)、新一代自适应统计迭代重建技术(ASIR)和基于模型的迭代重建算法(MBIR)三种图像重建技术对低剂量胸部薄层CT(LDCT)图像质量的影响。方法选取2020年6月1日中国医学科学院肿瘤医院收治的行LDCT扫描的连续20例无症状健康体检者,均为正常体型。采用FBP、50%ASIR和MBIR三种不同重建算法对LDCT原始数据行薄层重建,取主气道右锁骨下动脉(Layer 1)、主动脉弓上缘(Layer 2)、气管隆突(Layer 3)、右下肺静脉(Layer 4)和左膈顶(Layer 5)五个层面,测量不同重建算法下同层面薄层CT图像噪声值SD和大动脉与背部肌肉对比噪声比(CNR)。比较三种不同重建算法对同一组原始数据的重建时间。结果13例受检者检出非钙化结节,检出率65.0%。LDCT共检出肺内结节44枚,其中非钙化结节30枚(1.5枚/例)。其中实性结节23枚,亚实性结节7枚。对LDCT扫描后的原始数据进行FBP、50%ASIR、MBIR重建。其中FBP重建后5个不同层面及主气道内噪声值分别为31.6、21.2、23.0、26.1、36.9和18.3;50%ASIR重建后5个不同层面及主气道内噪声值分别为20.7、16.9、20.6、23.7、24.4和14.5;MBIR重建后5个不同层面及主气道内噪声值分别为14.3、16.2、18.4、17.6、18.5和14.2。与FBP重建图像比较,50%ASIR和MBIR平均噪声明显降低,组间差异均有统计学意义(均P <0.05)。在Layer 1和Layer 5层面,MBIR图像噪声较FBP下降54.7%和49.9%,较50%ASIR降低30.9%和24.2%。与FBP和50%ASIR相比,MBIR组图像的CNR明显提高(P <0.05)。MBIR在各层面的CNR分别为(76.6±26.8)、(76.3±26.4)、(76.4±26.4)、(76.2±25.9)和(75.9±26.2)。与FBP重建图像相比,CNR提高了32.2%,差异有统计学意义(P <0.05);与50%ASIR图像相比,CNR提高了5.1%(P> 0.05)。FBP、50%ASIR和MBIR三种不同重建算法重建LDCT原始数据的时间分别为(5.5±1.3) s、(6.3±2.2) s和(850.5±17.9) s,50%ASIR较FBP重建时间略长,MBIR组较其他两组延长(均P <0Objective To investigate the impact of three different reconstruction algorithms filtered back projection( FBP),adaptive statistical iterative reconstruction( ASIR) and model based iterative reconstruction( MBIR) on image quality of the low dose and thin slice computed tomographic( LDCT).Methods Twenty asymptomatic healthy individuals who underwent LDCT examination were enrolled on June 1,2020. All subjects had normal body shape. The raw data were reconstructed into thin slice using the three different reconstruction algorithms FBP,ASIR and MBIR. The standard deviation( SD) of noise level of a thin slice CT image and contrast to noise ratio( CNR) were estimated using the three different reconstruction algorithms at right subclavian arteries( Layer 1),aortic arch layer( Layer 2),carina of trachea layer( Layer 3),right lower pulmonary vein layer( Layer 4) and left parietal diaphragm layer( Layer 5),respectively. The reconstruction time of three different reconstruction algorithm on the same set of raw data was recorded and compared. Results Thirteen participants were diagnosed with non calcified lung nodules and the detection rate was 65. 0%. Fort-four pulmonary nodules were detected among which 30 pulmonary nodules were non calcified( 1. 5 per person) on LDCT. Of them,23 nodules were solid nodules( SNs) and 7 nodules were sub-solid nodules( SSNs). After LDCT scan,the raw data were reconstructed with MBIR,ASIR 50% and FBP. Noise level was 31. 6,21. 2,23. 0,26. 1,36. 9 and 18. 3,respectively for FBP,20. 7,16. 9,20. 6,23. 7、24. 4 and 14. 5,respectively for ASIR 50% and 14. 3,16. 2,18. 4,17. 6,18. 5 and 14. 2,respectively for MBIR at five layers. Compared with FBP and ASIR 50%,the noise level was low with MBIR( all P < 0. 05). At layer 1 and layer 5,the noise level decreased by 54. 7% and 49. 9%,respectively with MBIR compared with FBP and 30. 9% and 24. 2% compared with ASIR 50%. CNR increased with MBIR compared with MBIR and ASIR 50%( P < 0. 05). CNR was 76. 6 ± 26. 8,76. 3 ±26. 4,76. 4 ± 26. 4,76. 2 ± 25. 9 a

关 键 词：胸部 低剂量 计算机体层成像 迭代重建 图像质量 

分 类 号：R730.4[医药卫生—肿瘤]