DLIR联合MAR在膝关节置换术后能谱CT中的价值OA
The Value of Deep Learning Image Reconstruction Combined with Metal Artifact Reduction in Spectral CT for Postoperative Knee Arthroplasty
目的:探讨深度学习图像重建(DLIR)和去金属伪影技术(MAR)联合应用时,在能谱 CT图像中对膝关节置入物金属伪影的减少效果及相关影响.方法:本研究回顾性纳入30例膝关节置换术后患者的能谱CT数据,分别采用自适应统计迭代重建(50%ASiR-V)及 3种不同等级的深度学习图像重建(DLIR-L/M/H)进行图像重建,并对每种重建算法应用去金属伪影技术(MAR)再次重建.以 10 keV 为间隔,重建 40~140 keV(共 11个单能量)图像,最终获得4组(MAR联合4种重建算法)共计44个图像数据集.通过测量植入物旁感兴趣区CT值及标准差(SD值),计算伪影指数(AI)、信噪比(SNR)与对比噪声比(CNR)作为客观评估依据,同时对图像质量进行主观评分,以此完成图像质量评估.结果:客观评价结果显示,同一算法下 AI均随千电子伏特(keV)升高而下降,MAR-ASiR组 AI低于 MAR-DLIR各组.其中:130 keV 和 140 keV条件下 MAR-ASiR组金属伪影去除效果最佳;MAR-DLIR-H组在多个能级条件下,其 CNR值和 SNR值均优于 MAR-ASiR组,且110 keV条件下的软组织信噪比及组织对比度可达到最佳平衡.主观评价结果显示,MAR-DLIR-H 组图像质量评分高于 MAR-ASiR组.结论:DLIR重建算法联合 MAR技术重建的虚拟单能量图像(VMIs),其质量优于ASiR-V算法联合MAR技术重建的VMIs;110 keV 条件下的DLIR-H 重建算法得到的图像质量最佳.
Objective:To investigate the effectiveness and associated effects of combining deep learning image reconstruction(DLIR)with metal artifact reduction(MAR)to reduce metal artifacts from knee implants in spectral computed tomography(CT)images.Methods:This retrospective study included energy spectrum CT data from 30 patients who underwent knee arthroplasty.Image reconstruction was performed using adaptive statistical iterative reconstruction(50%ASiR-V)and three different levels of deep learning image reconstruction(DLIR-L/M/H).Subsequently,metal artifact reduction(MAR)was applied to each reconstruction to generate the second set of images.Images were reconstructed at energy levels ranging from 40 keV to 140 keV in 10 keV increments(11 single-energy levels in total),resulting in 44 image datasets across four groups(MAR combined with the four reconstruction algorithms).Image quality was assessed by measuring the CT number and standard deviation(SD)in the region of interest adjacent to the implant,as well as calculating the artifact index(AI),signal-to-noise ratio(SNR),and contrast-to-noise ratio(CNR)as objective evaluation criteria.Concurrently,we performed a subjective scoring of image quality to evaluate both image quality and diagnostic confidence.Results:The objective evaluation results revealed that using the same algorithm,AI decreased as keV increased.The AI in the MAR-ASiR group was lower than that in the MAR-DLIR groups,with the MAR-ASiR group demonstrating the best metal artifact removal performance at 130 keV and 140 keV.Across multiple energy levels,the MAR-DLIR-H group exhibited superior CNR and SNR values compared to the MAR-ASiR group,with an optimal balance between soft tissue signal-to-noise ratio and tissue contrast observed at 110 keV.The results of the subjective evaluation showed that the image quality and diagnostic confidence scores were higher in the MAR-DLIR-H group than in the MAR-ASiR group.Conclusion:VMI reconstructed using the DLIR algorithm combined with MAR exhibit superior quality compared to those reconstructed using the ASiR-V algorithm combined with MAR;the DLIR-H reconstruction algorithm yields the best image quality at 110 keV.
蔡琳沙;林晓娟;汤博荣;陈素萍;薛蕴菁;刘元芬
福建医科大学医学影像学院,福州 350122福建医科大学医学影像学院,福州 350122福建医科大学医学影像学院,福州 350122通用电气医疗CT影像研究中心,长沙 410000福建医科大学附属协和医院放射科,福州 350000福建医科大学附属协和医院放射科,福州 350000
信息技术与安全科学
深度学习图像重建去金属伪影技术膝关节置换术能谱CT
deep learning image reconstruction(DLIR)metal artifact reduction(MAR)total knee arthroplasty(TKA)energy-spectral CT
《CT理论与应用研究》 2026 (6)
1046-1054,9
评论