基于深度学习的图像重建算法在下肢动脉病变CTA诊断中的研究  

Study of Deep Learning-Based Image Reconstruction Algorithms in the Diagnosis of Lower Limb Arterial Lesions Using CTA

作　　者：陈芸[1] 朱彦[1] 王扬[2] 赵天[1] 李月峰[1] 陈兴兵 CHEN Yun;ZHU Yan;WANG Yang;ZHAO Tian;LI Yuefeng;CHEN Xingbing(Department of Radiology,Affiliated Hospital of Jiangsu University,Zhenjiang Jiangsu 212001,China;Department of Radiology,Gaoyou People’s Hospital,Yangzhou Jiangsu 225600,China)

机构地区：[1]江苏大学附属医院医学影像科,江苏镇江212001 [2]高邮市人民医院放射科,江苏扬州225600

出　　处：《中国医疗设备》2024年第3期134-138,共5页China Medical Devices

基　　金：江苏省重点研发计划(BE2017698)。

摘　　要：目的 探讨基于深度学习的图像重建算法对下肢动脉病变CT血管成像(Computed Tomography Angiography,CTA)的诊断价值。方法 回顾性收集2021年6月至2022年2月于我院就诊的51例下肢动脉狭窄或闭塞患者的CTA检查资料(65条下肢动脉)。分别基于深度学习的图像重建(Deep Learning Image Reconstruction,DLIR)算法和混合迭代重建(Hybrid Iterative Reconstruction,HIR)算法对CTA图像进行重建,以HIR法为参照进行质量评估;两位医师在不同重建算法下对血管狭窄的部位和程度进行评估,并采用Kappa检验观察者间一致性;以数字减影血管造影作为“金标准”比较HIR法和DLIR法诊断下肢动脉中度和重度狭窄的效能。结果与HIR法相比,DLIR法图像质量的噪声显著降低(Z膝上动脉=8.36,Z膝下动脉=9.46,Z足背动脉=7.19,均P<0.001),信噪比(Z膝上动脉=-7.32,Z膝下动脉=-7.91,Z足背动脉=-8.45,P<0.001)及对比噪声比(Z膝上动脉=-8.66,Z膝下动脉=-9.21,Z足背动脉=-8.52,均P<0.001)显著提高。DLIR法对动脉狭窄或闭塞程度的识别和评估均显示出更高的观察者间一致性(Kappa=0.86)。与HIR法相比,DLIR法的图像对膝下动脉重度狭窄的敏感度(72.2%vs.94.4%)、特异性(78.7%vs.95.7%),足背动脉中度狭窄的特异性(86.0%vs.97.7%)及重度狭窄的敏感度(50.0%vs.87.5%)均显著提高(P<0.05)。结论 DLIR算法可有效改善下肢动脉的CTA图像质量,并获得更优的诊断效能。Objective To explore the diagnostic value of deep learning based image reconstruction algorithms for computed tomography angiography(CTA)of lower limb arterial lesions.Methods CTA examination data from 51 patients(65 lower extremity arteries)with lower extremity arterial stenosis or occlusion who were treated in our hospital from June 2021 to February 2022 was retrospective collected.Based on the deep learning image reconstruction(DLIR)algorithm and hybrid iterative reconstruction(HIR)algorithm,CTA images were reconstructed separately,and the quality was evaluated using HIR as a reference.Two physicians assessed the location and degree of vascular stenosis under different reconstruction algorithms and observed interobserver consistency using Kappa test.Digital subtraction angiography was used as the“gold standard”to compare the performance of HIR and DLIR in diagnosing moderate and severe stenosis of lower extremity arteries.Results Compared with the HIR algorithm,the noise of image quality in DLIR algorithm was significantly reduced(ZSuperior knee artery=8.36,ZInfrapopliteal artery=9.46,ZDorsalis pedis artery=7.19,P<0.001),and signal to noise ratio(ZSuperior knee artery=-7.32,ZInfrapopliteal artery=-7.91,ZDorsalis pedis artery=-8.45,P<0.001)and contrast to noise ratio was significantly improved(ZSuperior knee artery=-8.66,ZInfrapopliteal artery=-9.21,ZDorsalis pedis artery=-8.52,P<0.001).Compared with the HIR method,images reconstructed based on DLIR showed significantly improved sensitivity(72.2%vs.94.4%)and specificity(78.7%vs.95.7%)for severe stenosis of the inferior knee artery,specificity for moderate stenosis in the dorsal foot artery(86.0%vs.97.7%)and the sensitivity for severe stenosis(50.0%vs.87.5%)(P<0.05).Conclusion DLIR algorithm can effectively enhance the quality of CTA images of lower extremity arteries,leading to improved diagnostic efficiency.

关 键 词：下肢动脉 深度学习 混合迭代重建 计算机断层扫描血管造影 数字减影血管造影 

分 类 号：R816.2[医药卫生—放射医学]