肠系膜下动脉CTA应用深度学习重建算法降低辐射剂量的可行性研究  

A Feasibility Study of Realizing Low-Dose CT Angiography of Inferior Mesenteric Artery Using Deep Learning Reconstruction Algorithm

作　　者：张竞颐 刘义军[1] 王诗耕 童小雨 范勇 陈安良[1] 胡梦婷 程启烨 ZHANG Jingyi;LIU Yijun;WANG Shigeng;TONG Xiaoyu;FAN Yong;CHEN Anliang;HU Mengting;CHENG Qiye(Department of Radiology,The First Affiliated Hospital of Dalian Medical University)

机构地区：[1]辽宁省大连市大连医科大学附属第一医院放射科

出　　处：《中国医学计算机成像杂志》2024年第5期616-621,共6页Chinese Computed Medical Imaging

摘　　要：目的:探究不同等级的深度学习重建算法(DL)对低剂量肠系膜下动脉(IMA)CT血管成像(CTA)图像质量的影响,并获得IMACTA图像最佳DL等级。方法:前瞻性收集行IMACTA的患者60例,随机分为A组和B组。A组(n=30)管电压120kV,管电流为剂量调制等级3,重建Karl5图像;B组(n=30)管电压120kV,管电流为剂量调制等级2,重建DL(1~4)等级图像,记为B1~B4组,并记录2组的辐射剂量。在轴位图像上测量IMA主干、同层腹主动脉和竖脊肌的CT值和噪声(SD)值,计算各血管的信噪比(SNR)对比度噪声比(CNR)和质量因数(FOM)。重组容积再现(VR)和最大密度投影(MIP)图像,由2名诊断医师对二维和三维图像进行5分制评分。结果:B组有效辐射剂量(ED)与A组相比降低了42.08%(P<0.05)。随着DL等级的提高,B1~B4组SD值逐渐降低,SNR,CNR和FOM逐渐提高(P<0.05)。B3、B4组腹主动脉的SNR、CNR、FOM优于A组(P<0.05)。2名诊断医师对图像主观评分一致性良好(Kappa值0.795~0.891,P<0.05)。B2、B3组与A组主观评分差异均无统计学意义(均P>0.05)。结论:深度学习重建算法在保证IMACTA图像质量的前提下实现了降低辐射剂量的目的。在满足诊断需求的情况下,DL3为临床推荐的最优重建等级。Purpose:To explore the effect of different levels of deep learning reconstruction algorithm(DL)on the image quality of low-dose CT angiography(CTA)of inferior mesenteric artery(IMA),as well as to find the most recommended level.Methods:A total of 60 patients who underwent CTA of IMA were prospectively collected and randomly assigned to groups A and B.The scanning and reconstruction parameters for group A(n=30)were 120 kV,tube current modulation 3 and Karl 5,whereas these for group B(n=30)were 120 kV,tube current modulation 2 and DL level 1-4(B1-B4).The radiation dose of two groups was also recorded.The CT and standard deviation(SD)values of IMA,abdominal aorta and erector spinae were measured to calculate the signal-to-noise ratio(SNR),contrast-to-noise ratio(CNR)and figure of merit(FOM).The volume rendering(VR)and maximum intensity projection(MIP)images were reconstructed,and two radiologists used a five-point scale to evaluate two-dimensional and three-dimensional images.Results:In comparison with group A,the effective dose(ED)of group B decreased by 42.08%(P<0.05).As the strength of DL increased,the SD values from B1 to B4 decreased gradually,whereas SNR,CNR and FOM increased gradually(P<0.05).The SNR,CNR and FOM of abdominal aorta in group B3 and B4 were superior than those in group A(P<0.05).Two radiologists had good inter-reader consistency(Kappa value:0.795-0.891,P<0.05).Compared with group A,group B2 and B3 had no statistically significant variation in the subjective evaluation(P>0.05).Conclusions:Deep learning reconstruction algorithm can ensure the image quality of IMA CTA and realize the purpose of decreasing radiation dose.Under the condition of satisfying the diagnostic requirement,DL 3 is the most optimal reconstruction strength for clinical recommendation.

关 键 词：辐射剂量 深度学习算法 肠系膜下动脉 图像质量 

分 类 号：R812[医药卫生—放射医学]