胃癌双能量CT低keV深度学习重建图像与常规图像对比研究  

Comparative study of low-keV deep learning reconstructed images and conventional images of gastric cancer based on dual-energy CT

作　　者：袁梦晨 刘译阳 梁何俊 陈林 赵帅 尤亚茹 高剑波[1] Yuan Mengchen;Liu Yiyang;Liang Hejun;Chen Lin;Zhao Shuai;You Yaru;Gao Jianbo(Department of Radiology,the First Affiliated Hospital of Zhengzhou University,Henan International Joint Laboratory of Medical Imaging,Henan Key Laboratory of Image Diagnosis and Treatment for Digestive System Tumor,Zhengzhou 450052,China;Department of Gastroenterology,Xuanwu Hospital,Capital Medical University,Beijing 100053,China)

机构地区：[1]郑州大学第一附属医院放射科、河南省医学影像国际联合实验室河南省消化肿瘤影像重点实验室,郑州450052 [2]首都医科大学宣武医院消化科,北京100053

出　　处：《中华放射学杂志》2024年第8期836-842,共7页Chinese Journal of Radiology

基　　金：国家自然科学基金(81971615)。

摘　　要：目的将深度学习图像重建(DLIR)算法与双能量CT(DECT)结合,评估胃癌低keV单能图像质量,并与常规自适应统计迭代重建(ASiR-V)算法图像进行比较。方法该研究为横断面研究,前瞻性收集2022年9月至2023年3月郑州大学第一附属医院31例胃癌患者的DECT图像,基于动脉期和静脉期图像,使用DLIR算法分别在低、中、高强度水平下(DLIR-L、DLIR-M、DLIR-H)重建55 keV单能量图像,同时使用ASiR-V算法重建70 keV 40%混合系数(ASiR-V40%)图像。在图像客观评价中,分别在4组重建图像上计算病灶和肌肉的信噪比(SNR)、对比噪声比(CNR)。在图像的主观评价中,对各组重建图像的总体图像质量、病灶显示能力及诊断信心进行评分。4组间SNR和CNR的比较采用单因素重复测量方差分析或Friedman检验,评分的比较采用Friedman检验,用Bonferroni校正对事后两两比较进行调整。结果在客观评价中,动脉期和静脉期55 keV DLIR-H图像上CNR病灶、SNR病灶以及SNR肌肉最高,且在70 keV ASiR-V40%、55 keV DLIR-L、DLIR-M以及DLIR-H图像上呈逐渐升高的趋势(P<0.05)。在主观评价中,与70 keV ASiR-V40%相比,55 keV DLIR-H的总体图像质量评分升高,但差异无统计学意义(P>0.05),DLIR-M评分相似或略差,而DLIR-L图像质量评分下降(P<0.05)。55 keV DLIR的各水平重建图像的病灶显示能力及诊断信心在动、静脉期均高于70 keV ASiR V40%图像(P<0.05)。结论相比70 keV ASiR-V40%常规图像,55 keV DLIR-H图像具有更高的病灶对比度和诊断信心,且图像噪声更低。55 keV DLIR-M图像与70 keV ASiR-V40%的整体图像质量相当,但前者的病灶对比度和诊断信心更高。55 keV DLIR-L无法将图像质量提升至70 keV ASiR-V40%水平。Objective To assess the quality of low-keV monoenergetic images using deep learning image reconstruction(DLIR)algorithm combined with dual energy CT(DECT)in gastric cancer and to compare them with images from the conventional adaptive statistical iterative reconstruction(ASiR-V)algorithm.Methods In this cross-sectional study,DECT images of 31 gastric cancer patients in the First Affiliated Hospital of Zhengzhou University were prospectively collected from September 2022 to March 2023.The 55 keV monoenergy images were reconstructed using the DLIR algorithm at low-,medium-,and high-intensity levels(DLIR-L,DLIR-M,and DLIR-H)based on arterial phase and venous phase images,respectively.The 70 keV 40%mixing coefficient(ASiR-V40%)images were reconstructed using the ASiR-V algorithm.In the objective evaluation of images,the signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)for both lesions and muscle were calculated across four sets of reconstructed images.In the subjective evaluation of images,scores were assigned to the overall image quality,lesion visibility,and diagnostic confidence for each set of reconstructed images.Comparisons of SNR and CNR between the 4 groups were made by One-way repeated-measures ANOVA or Friedman′s test.Comparisons of scores were made by Friedman′s test.The P value of pairwise comparison was adjusted using Bonferroni correction methods.Results In the objective evaluations,CNR lesion,SNR lesion and SNR muscle were highest on the 55 keV DLIR-H images in the arterial and venous phases,and showed a gradually increasing trend on the 70 keV ASiR-V40%,55 keV DLIR-L,DLIR-M,DLIR-H images(P<0.05).In subjective evaluations,compared to the 70 keV ASiR-V40%images,overall image quality scores were numerically higher for the 55 keV DLIR-H(P>0.05),similar or slightly worse for the 55 keV DLIR-M,and significantly lower for the 55 keV DLIR-L(P<0.05).The lesion visibility and diagnostic confidence on the 55 keV DLIR reconstruction images were higher in both arterial and venous phases than 70 keV AS

关 键 词：胃肿瘤 体层摄影术 X线计算机 深度学习 图像重建 

分 类 号：R735.2[医药卫生—肿瘤] R730.44[医药卫生—临床医学]