深度学习图像重建联合去金属伪影算法在颌面部CT图像的应用  

Application of deep learning image reconstruction combined with metal artifact reduction algorithm in maxillofacial CT images

作　　者：唐丽[1] 魏一娟[1] 侯平[1] 查开继[1] 高剑波[1] TANG Li;WEI Yijuan;HOU Ping;ZHA Kaiji;GAO Jianbo(Department of Radiology,the First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China)

机构地区：[1]郑州大学第一附属医院放射科,河南郑州450052

出　　处：《实用放射学杂志》2024年第8期1363-1366,共4页Journal of Practical Radiology

摘　　要：目的探讨深度学习图像重建(DLIR)联合Smart去金属伪影(Smart MAR)算法在颌面部CT图像的应用价值。方法纳入行颌面部增强CT检查且颌面部病变受口腔金属植入物干扰的患者34例。采用静脉期分别重建4组图像:50%自适应统计迭代重建(ASIR-V)(IR组)、50%ASIR-V联合Smart MAR(IR+S组)、中档DLIR联合Smart MAR(D-M+S组)及高档DLIR联合Smart MAR(D-H+S组)。通过测量颌面部病变及头长肌CT值的标准差(SD)计算伪影指数(AI);对各组图像整体图像质量、病灶显著性和诊断信心进行主观评分。比较不同重建算法对图像质量的影响。结果IR组的AI值较IR+S组升高(P<0.05),而2组的噪声差异无统计学意义(P>0.05)。D-M+S组及D-H+S组AI值及噪声较IR+S组均降低(P<0.05),AI值分别下降13.70%、19.06%,噪声分别下降16.37%、30.78%。IR+S组在整体图像质量、病灶显著性和诊断信心方面主观评分均低于D-M+S组及D-H+S组,高于IR组(P<0.05)。6例患者(17.64%)仅在加入Smart MAR算法组中检测到病变。然而,9例患者(26.47%)在Smart MAR算法组中舌部出现新的伪影。结论DLIR联合Smart MAR可显著改善口腔金属植入物患者颌面部CT的图像质量、提高颌面部病变的显著性和诊断信心。Smart MAR算法可能产生新的伪影,需与无Smart MAR算法的图像一起分析。Objective To explore the application value of deep learning image reconstruction(DLIR)combined with Smart metal artifact reduction(Smart MAR)algorithm in maxillofacial CT images.Methods A total of 34 patients with maxillofacial lesions affected by oral metal implants who underwent maxillofacial enhanced CT scans were included.The images of four groups in venous phase were reconstructed with 50%adaptive statistical iterative reconstruction(ASIR-V)(IR group),50%ASIR-V combined with Smart MAR(IR+S group),DLIR(at medium strength)combined with Smart MAR(D-M+S group)and DLIR(at high strength)combined with Smart MAR(D-H+S group)respectively.The artifact index(AI)was worked out by measuring the standard deviation(SD)of CT values in maxillofacial lesions and longhead muscle.The subjective scores of overall image quality,lesion conspicuity and diagnostic confidence were assessed.The image quality of different algorithms was compared.Results Compared with IR+S group,the AI value of IR group was significantly increased(P<0.05),while the noise had no significant difference(P>0.05).Compared with IR+S group,the AI value and noise of D-M+S group and D-H+S group both were significantly decreased(P<0.05),and the AI value of D-M+S group and D-H+S group reduced by 13.70%and 19.06%respectively,the noise reduced by 16.37%and 30.78%respectively.The subjective scores of overall image quality,lesion conspicuity and diagnostic confidence in IR+S group were significantly lower than those in D-M+S group and D-H+S group,but significantly higher than those in IR group(P<0.05).There were 6 patients’(17.64%)lesions were detected only in the groups with Smart MAR algorithm,while 9 patients(26.47%)had introduced new artifacts in the tongue with Smart MAR algorithm.Conclusion DLIR combined with Smart MAR can improve the CT image quality of maxillofacial region,enhance the conspicuity and diagnosis confidence of maxillofacial lesions in patients with oral metal implants.Smart MAR algorithm may produce new artifacts that need to be analyzed along

关 键 词：计算机体层成像 深度学习图像重建 金属伪影 颌面部 

分 类 号：R814.42[医药卫生—影像医学与核医学] R816.2[医药卫生—放射医学]