深度学习重建算法对低管电压、低对比剂用量的头颈CTA图像质量的影响  

Impact of deep learning reconstruction algorithm on the image quality of head and neck CT angiography with low tube voltage and low contrast agent dosage

作　　者：张欣玥 王沄[1] 陈钰[1] 童佳静 苏童 陆晓平 徐敏 王剑 张竹花[1] 金征宇[1] ZHANG Xinyue;WANG Yun;CHEN Yu;TONG Jiajing;SU Tong;LU Xiaoping;XU Min;WANG Jian;ZHANG Zhuhua;JIN Zhengyu(Department of Radiology,Peking Union Medical College Hospital,Chinese Academy of Medical Sciences&Peking Union Medical College,Beijing 100730,China)

机构地区：[1]中国医学科学院北京协和医学院北京协和医院放射科,北京100730 [2]佳能医疗系统(中国)有限公司,北京100027

出　　处：《影像诊断与介入放射学》2025年第1期3-8,共6页Diagnostic Imaging & Interventional Radiology

基　　金：国家自然科学基金(82001814)。

摘　　要：目的研究深度学习重建算法(DLR)对低管电压、低对比剂用量的头颈CT血管造影(CTA)图像质量的影响。方法前瞻性收集我院2022年4月—2023年11月行头颈CTA检查的患者66例,并随机分为两组。实验组采用80 kVp管电压,注射对比剂碘帕醇(370 mg I/mL)28 mL,生理盐水40 mL,采用混合迭代重建(HIR)和DLR的低、中、高强度算法重建出4组图像,分别命名为L-HIR组、L-DLR_(mild)组、L-DLR_(standard)组、L-DLR_(strong)组。对照组采用100 kVp管电压,注射对比剂40 mL,生理盐水40 mL,采用HIR重建,命名为R-HIR组。测量并计算主动脉弓、颈内动脉、颈外动脉、椎动脉、基底动脉、大脑中动脉平均CT值、图像噪声(SD)、信噪比(SNR)和对比噪声比(CNR)等客观指标,并对5组图像进行5分制主观评分。根据数据是否满足正态分布,选择采用独立样本的t检验或者Mann-Whitney U检验来比较各组的图像质量。结果3组L-DLR图像在主动脉弓、基底动脉、椎动脉、颈内动脉的CT值均大于R-HIR组(P<0.05),L-DLR_(mild)组在主动脉弓、颈内动脉层面的SD值与R-HIR组无显著差异(P>0.05),其余L-DLR组的SD值均小于R-HIR(P<0.05),L-DLR_(standard)组和L-DLR_(strong)组的SNR值均大于R-HIR(P<0.05),L-DLR_(strong)组的CNR值均大于R-HIR(P<0.001)。5组图像的主观评分分别为:R-HIR(3.93±0.27)、L-HIR(2.72±0.82)、L-DLR_(mild)(3.34±0.82)、L-DLR_(standard)(3.52±0.6)、L-DLR_(strong)(3.61±0.79),除L-HIR组外均符合诊断要求。实验组有效辐射剂量较对照组减少了65%[(0.41±0.08)mSv比(1.18±0.12)mSv]。结论采用80 kVp管电压和28 mL对比剂,结合深度学习重建算法可在客观上获得与100 kVp管电压和40 mL对比剂相当的头颈CTA图像质量。该方法减少辐射剂量和对比剂用量,提高头颈CTA检查的安全性。Objective To investigate the effect of deep learning reconstruction(DLR)algorithm on the image quality of head and neck CT angiography(CTA)with low tube voltage and low contrast agent dosage.Methods A total of 66 patients who underwent head and neck CTA in our hospital from April 2022 to November 2023 were prospectively enrolled and randomly divided into two groups.CTA of the experimental group was performed using 80 kVp tube voltage,28 mL of contrast agent,and 40 mL of saline.Images were reconstructed with hybrid iterative reconstruction(HIR),DLR at mild-,standard-and strong-strength producing four datasets of low-dose HIR(L-HIR),L-DLR_(mild),L-DLR_(standard),and L-DLR_(strong).CTA of the control group was performed using 100 kVp tube voltage,40 mL of contrast agent,and 40 mL of saline with HIR producing regular-dose HIR dataset(R-HIR).Objective indexes such as aortic arch,internal carotid artery,external carotid artery,vertebral artery,basilar artery,middle cerebral artery,image noise(standard deviation,SD),signal-to-noise ratio(SNR),and contrast-to-noise ratio(CNR)were determined and the five groups of images were subjectively scored.The independent samples t-test or Mann-Whitney U test was used to compare the image quality between the two groups.Results The CT values of three L-DLR image groups in the aortic arch,basilar artery,vertebral artery,and internal carotid artery were all significantly higher than those of the R-HIR group(P<0.05).The L-DLR_(mild)group showed no significant difference at the aortic arch and internal carotid artery levels from that of the R-HIR group(P>0.05)whereas the SD values were significantly lower and SNR significantly higher in the L-DLR_(standard)and L-DLR_(strong)groups than those of R-HIR group(P<0.05).The CNR of the L-DLR_(strong)group was significantly higher than that of the R-HIR group(P<0.001).The subjective scores of R-HIR(3.93±0.27),L-DLR_(mild)(3.34±0.82),L-DLR_(standard)(3.52±0.6),and L-DLR_(strong)(3.61±0.79)met diagnostic requirements whereas the L-HIR score(2.

关 键 词：CT血管造影 深度学习重建 辐射剂量 图像质量 

分 类 号：R814.43[医药卫生—影像医学与核医学] R144.1[医药卫生—放射医学]