人工智能迭代重建算法对超低剂量胸部CT图像质量和计算机辅助肺结节检测的影响  

Impacts of artificial intelligence iterative reconstruction algorithm on the image quality and computer-aided pulmonary nodule detection at ultra-low dose chest CT

作　　者：张宝平 李傲 李宇航 朱书萌 田倩 赵文哲 肖瑶 侯伟 刘哲[1] 王睿 黄欣[1] 郝辉[1] 王怡名 杨健[1,2] 金超 Zhang Baoping;Li Ao;Li Yuhang;Zhu Shumeng;Tian Qian;Zhao Wenzhe;Xiao Yao;Hou Wei;Liu Zhe;Wang Rui;Huang Xin;Hao Hui;Wang Yiming;Yang Jian;Jin Chao(Department of Radiology,The First Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710061,China;Shaanxi Engineering Research Center of Computational Imaging and Medical Intelligence,Xi’an 710061,China)

机构地区：[1]西安交通大学第一附属医院医学影像科,陕西西安710061 [2]陕西省计算影像与医疗智能工程研究中心,陕西西安710061

出　　处：《遵义医科大学学报》2024年第4期384-391,共8页Journal of Zunyi Medical University

基　　金：国家自然科学基金数学天元重点专项(NO:12226007)。

摘　　要：目的 探讨人工智能迭代重建(AIIR)算法对超低剂量胸部CT图像质量和计算机辅助(CAD)肺结节检测的影响。方法 前瞻性纳入2023年9月至10月于我院行胸部CT检查的患者41例,同时采集常规剂量(120 kVp),采用管电流调制技术,剂量等级2级,参考管电流(106 mAs)和超低剂量(120 kVp,15 mAs)胸部CT,记录辐射剂量参数。常规剂量CT采用Karl迭代重建,超低剂量CT采用Karl和AIIR(1、3、5级)重建,测量5组图像的主动脉、脂肪、肌肉区域的CT均值、噪声指数(SD)并计算信噪比(SNR)和对比噪声比(CNR),主观评价5组图像的整体图像质量。以2名胸部影像诊断医师共同阅片确定的≥4 mm非钙化实性肺结节作为参考标准,记录5组图像CAD肺结节检测的真阳性、假阳性(误诊)和假阴性(漏诊)数量并与参考标准比较,记录超低剂量Karl重建组图像人工阅片时间和CAD阅片时间。结果 5组图像CT值差异无统计学意义(P>0.05),超低剂量AIIR 5重建的SD、SNR和CNR值均优于常规剂量和超低剂量Karl重建,差异均有统计学意义(P<0.005)。超低剂量AIIR 5重建的胸部CT整体图像质量与常规剂量Karl重建相当,优于超低剂量Karl重建(P<0.005)。5组图像CAD肺结节检测的灵敏度差异无统计学意义(P>0.05),超低剂量AIIR重建CAD肺结节检测的假阳性率低于常规剂量和超低剂量Karl重建,差异均有统计学意义(P<0.005)。超低剂量组辐射剂量较常规剂量组降低约89.3%,CAD阅片时间较人工阅片减少63.3%。结论 AIIR在改善超低剂量胸部CT图像质量的同时可以保持CAD肺结节检测的灵敏度,降低CAD肺结节检测的假阳性。Objective To investigate the effect of artificial intelligence iterative reconstruction(AIIR) algorithm on the image quality and computer-aided pulmonary nodule detection at ultra-low dose chest CT.Methods Forty-one patients who underwent chest CT examination in our hospital from September to October 2023 were prospectively enrolled.Conventional dose(120 kVp,using tube current modulation technology,dose level 2,reference tube current 106 mAs) and ultra-low dose(120 kVp,15 mAs) chest CT were simultaneously collected,the radiation dose parameters were recorded.Conventional dose CT was reconstructed by Karl iterative reconstruction and ultra-low dose CT was reconstructed by Karl and AIIR(level 1,3,and 5).The CT value and noise index(Standard deviation,SD) of the aorta,fat and muscle areas in the five groups of images were measured and the signal to noise ratio(SNR) and contrast to noise ratio(CNR) were calculated.The overall image quality of the five groups of images was subjectively evaluated.Using≥4 mm non-calcified solid pulmonary nodules determined by two chest radiologists who jointly reviewed the images as the reference standard,the true positive,false positive(misdiagnosis),and false negative(missed diagnosis) pulmonary nodule numbers of the five groups of images CAD pulmonary nodule detection were recorded and compared with the reference standard.The manual review time and CAD review time of the ultra-low dose Karl reconstruction group were recorded.Results There was no significant difference in CT value among the five groups(P>0.05).The SD,SNR and CNR values of ultra-low dose AIIR reconstruction were better than those of conventional dose and ultra-low dose Karl reconstruction,with statistically significant differences(P<0.005).The overall image quality of ultra-low dose AIIR5 reconstruction was similar to that of conventional dose Karl reconstruction and was better than that of ultra-low dose Karl reconstruction(P<0.005).There was no significant difference in sensitivity of pulmonary nodule detection amo

关 键 词：胸部CT 肺结节 人工智能迭代重建 超低剂量CT 

分 类 号：R455.3[医药卫生—运动医学]