深度学习重建算法对低kV逆血流扫描下肢动脉CT血管成像图像质量的影响  

The effects of deep-learning image reconstruction algorithm on image quality of lower extremity CT angiography with low kV and reverse flow direction scanning

作　　者：陈依林 刘元芬[1] 王莉莉[1] 叶雄鑫 薛蕴菁[1] Chen Yilin;Liu Yuanfen;Wang Lili;Ye Xiongxin;Xue Yunjing(Department of Radiology,Union Hospital Affiliated to Fujian Medical University,Fuzhou 350001,China)

机构地区：[1]福建医科大学附属协和医院放射科,福州350001

出　　处：《中华放射学杂志》2022年第11期1188-1194,共7页Chinese Journal of Radiology

摘　　要：目的探讨深度学习图像重建(DLIR)算法在低kV逆血流扫描下肢动脉CT血管成像(CTA)中对图像质量的影响。方法回顾性收集2021年1月至3月50例于福建医科大学附属协和医院因临床可疑下肢动脉疾病而行低kV逆血流扫描下肢动脉CTA患者的CT原始数据, 分别使用具有3种混合权重的自适应统计迭代重建(ASIR-V)算法(ASIR-V 20%、ASIR-V 50%、ASIR-V 80%)、3种重建强度(高、中、低)的DLIR算法(DLIR-H、DLIR-M、DLIR-L)在0.625 mm上重建薄层轴面图像。在每组图像上测量近端腹主动脉、远端腹主动脉、左右髂总动脉、左右股动脉(上段)、左右股浅动脉(中段)、左右腘动脉的CT值、噪声(SD)值, 并计算信噪比(SNR)和对比噪声比(CNR)。由两名医师采用4分法将下肢血管分成4个节段对图像的噪声和锐利度进行主观评价。统计学分析采用单因素方差分析评估图像质量主观评分及客观评价指标的差异。结果随着ASIR-V重建权重的增加和DLIR重建强度的增加, 各测量层面的SD值逐渐降低, SNR值和CNR值逐渐增高(P均<0.05)。6种重建方式中, DLIR-H重建与ASIR-V80%重建具有较低的SD值及较高的SNR与CNR值(P均<0.05)。与ASIR-V20%和ASIR-V50%相比, DLIR-H组具有更低的SD值, 更高的SNR和CNR值(P均<0.05);但DLIR-H组与ASIR-V80%相比, SD、SNR和CNR值的差异无统计学意义(P均>0.05)。主观评分结果显示, 在腹主动脉-足部动脉全程, DLIR-H组噪声评分为最优, 锐利度评分也明显高于ASIR-V80%(P均<0.05)。结论深度学习重建可以显著降低下肢动脉CTA图像的噪声, 提升图像质量。在DLIR(H、M、L)3种深度学习重建算法中, DLIR-H的降噪效能以及在图像噪声和锐利度的平衡中表现最佳, 具有最高的图像质量。Objective To investigate impacts of a deep learning image reconstruction(DLIR)algorithm on image quality of lower extremity CTA with low kVp and reverse flow direction scanning.Methods From January 2021 to March 2021,fifty patients with suspected lower extremities diseases and received lower extremity CTA with low kVp and reverse flow direction scanning in Union Hospital affiliated to Fujian Medical University were retrospectively collected in this study.Six groups of CT images were reconstructed at the thickness of 0.625 mm using two algorithms including ASIR-V of three blending ratios(ASIR-V 20%,ASIR-V 50%and ASIR-V 80%)and DLIR of three strengths(DLIR-H,DLIR-M and DLIR-L).Regions of interest(ROIs)were placed on proximal abdominal aorta(AA),distal AA,left and right common iliac arteries,left and right femoral arteries(upper segment),left and right superficial femoral arteries(middle segment),left and right popliteal arteries.The CT value and SD value were measured for each group;the signal-noise ratio(SNR)and contrast-noise ratio(CNR)were calculated.The lower extremity CTA was divided into four segments,and the subjective evaluation was independently performed on noise and sharpness using 4 points scales by two radiologists.One-way analysis of variance was utilized to evaluate the differences in subjective scoring and objective parameters among the six groups.Results For all arteries segments,with the increase of blending ratios for ASIR-V and reconstruction strength of DLIR,the SD values were reduced while SNR and CNR were increased(all P<0.05).Among the six groups,DLIR-H and ASIR-V80%images had lowest SD as well as highest SNR and CNR(all P<0.05).In comparison to ASIR-V20%and ASIR-V50%images,DLIR-H images showed lower SD,higher SNR and CNR values(all P<0.05).There were no statistical differences between ASIR-V80%and DLIR-H images in SD,SNR and CNR values(all P>0.05).Subjective scoring results showed that the DLIR-H images displayed the best noise performance for the entire lower extremity arteries from AA to

关 键 词：体层摄影术 X线计算机 人工智能 深度学习 下肢动脉 图像质量 

分 类 号：R816.2[医药卫生—放射医学] R543.5[医药卫生—临床医学]