摘要
目的:通过与滤波反投影(FBP)算法和自适应统计迭代重建(ASIR-V)算法对比,探讨深度学习图像重建(DLIR)算法对肺部CT定量分析及图像质量的影响。方法:回顾性搜集46例肺部体检患者的CT平扫数据,采用FBP、不同混合权重(BW=50%、100%)ASIR-V算法及不同级别深度学习迭代重建(DLIR-L、DLIR-M、DLIR-H)算法进行图像重建。测量并比较不同重建算法图像上各结构及病灶的定量参数,包括肺结节的平均CT值及实性和非实性部分的体积、全肺容积、肺气肿指数(EI)、右肺上叶尖段支气管的气道面积和平均直径、气道壁的面积、面积百分比和平均厚度、气管分叉层面气道内空气和降主动脉CT值及其标准差(图像噪声)。由两位放射科医师从图像质量及噪声水平两方面对不同重建算法图像采用5分法(1分:极差,2分:差,3分:满足诊断要求,4分:好,5分:极佳)进行主观评价并进行统计学分析。测量数据符合正态分布的连续变量采用单因素方差分析,不符合者采用非参数检验,P<0.05时,进一步组间两两比较,采用Kappa检验比较两位医师主观评分的一致性。结果:与FBP算法相比,不同级别DLIR及50%和100%权重ASIR-V算法可显著降低图像噪声(P均<0.001)。与FBP、50%BW-ASIR-V、DLIR-L和DLIR-M算法相比,DLIR-H算法可显著降低图像噪声(P均<0.001)。不同算法对肺气肿指数测量值具有显著影响(P<0.001)。不同算法对全肺容积、右肺上叶尖段支气管的气道腔面积、气道壁面积、气道壁平均厚度和气道平均直径、气道内空气及降主动脉CT值及结节体积的测量结果无显著影响(P均>0.05)。DLIR-H算法图像质量主观评分高于其它算法(P均<0.05),图像噪声的主观评分高于FBP、50%-ASIR-V、DLIR-L及DLIR-H(P均<0.05)。结论:与FBP算法相比,DLIR算法和ASIR-V算法能够在降低图像噪声和提高图像质量的同时不影响肺部各结构及肺结节定量指标的测量结果;与AS
Objective:This study was aimed to investigate the impact of deep learning image reconstruction(DLIR)algorithm on quantitative analysis and image quality of pulmonary CT by comparing with filtered back projection(FBP)and adaptive statistical iterative reconstruction veo(ASIR-V).Methods:Non-contrastimages of 46 subjects who underwent lung CT screeningwere collected retrospectively and reconstructed using FBP,ASIR-V with 50%and 100%hybrid iterative reconstruction blending weights(BW),and different levels(low,medium and high)of deep learning-based reconstruction algorithm(named DLIR-L,DLIR-M and DLIR-H,respectively).Quantitative parameters were compared including average CT values and volumes of solid(VS)and non-solid areas(VNS)of each nodule,lung volume(LV),emphysema index(EI),and luminal area(LA),airway diameter(AD),wall area(WA)and wall thickness(WT)of the right upper lobular apical segment,CT value and its standard deviation(SD,represented as image noise)of the descending aorta and CT value and its SD of the air inside the trachea at the level 1cm above the tracheal bifurcation.Image quality and noise in different reconstruction algorithms were evaluated by two radiologists using a 5-point scale(1:very poor;2:poor;3:qualified;4:good;5:excellent)and compared statistically.Results:Compared with FBP,the image noise was significantly reducedby using different level DLIR algorithms and ASIR-V algorithms with 50%and 100%BW(all P<0.001).Compared with FBP,the image noise was significantly reduced by using 50%-ASIR-V,DLIR-L and DLIR-M algorithms,DLIR-H algorithms(P<0.001).Different algorithms had a significant effect on emphysema index(P<0.001).There were no significant differences in total lung volume,airway area,airway wall area,average airway wall thickness,airway diameter,CT values of air and descending aorta,and nodule volume(VS and VNS)among different reconstructions(all P>0.05).DLIR-H algorithm had higher subjective evaluating score of image quality than that of the other algorithms(all P<0.05).And the subjective s
作者
张卓璐
陈菁
刘卓
陈雷
洪楠
ZHANG Zhuo-lu;CHEN Jing;LIU Zhuo(Department of Radiology,People’s Hospital,Peking University,Beijing 100044,China)
出处
《放射学实践》
CSCD
北大核心
2023年第4期434-440,共7页
Radiologic Practice
关键词
迭代重建
深度学习
图像重建
体层摄影术
X线计算机
肺结节
肺气肿
Iterative reconstruction
Deep learning
Image reconstruction
Tomography,X-ray computed
Pulmonary nodule
Pulmonary emphysema
