深度学习图像重建算法对改善直肠CT图像质量的临床应用价值  

Improving rectal CT image quality with a deep learning image reconstruction algorithm

作　　者：乔文俊[1,2] 周芳 刘泉芬[1] 黄婵桃 许乙凯[1,2] QIAO Wenjun;ZHOU Fang;LIU Quanfen;HUANG Chantao;XU Yikai(Department of Imaging Diagnostics,Nanfang Hospital,Southern Medical University,Guangzhou 510515,China;Medical Radiation Protection Committee of Radiation Protection Association of Guangdong Province,Guangzhou 510515,China)

机构地区：[1]南方医科大学南方医院影像诊断科,广东广州510515 [2]广东省辐射防护协会医学辐射防护专委会,广东广州510515

出　　处：《中国医学物理学杂志》2024年第8期975-981,共7页Chinese Journal of Medical Physics

基　　金：国家自然科学基金(82371655)。

摘　　要：目的:探索深度学习图像重建(DLIR)算法是否可以改善静脉期肛管直肠的CT图像质量。方法:回顾性纳入进行腹部CT增强扫描的71例患者,所有影像资料使用50%ASiR-V和DLIR低、中、高(DLIR-L、DLIR-M、DLIR-H)3个强度的DLIR重建静脉期薄层图像。测量各组图像的肛管和臀部脂肪的CT值和标准差(SD),以臀部脂肪SD作为背景噪声,计算肛管对比噪声比(CNR)和信噪比(SNR)。两名影像科医师使用Likert5分量表法独立进行图像质量评估和直肠癌局部侵犯情况诊断信心评价。分析比较客观测量指标和图像主观评分,采用Kappa检验评估一致性。结果:各组间肛管CT值及臀部脂肪CT值比较差异没有统计学意义(P>0.05),脂肪SD、肛管SNR及CNR比较差异有统计学意义(P<0.05),DLIR-H组脂肪SD最低,SNR及CNR最高,而50%ASiR-V组脂肪SD最高,SNR及CNR最低。与50%ASiR-V组相比,DLIR-H组脂肪SD降低44.3%,肛管SNR及CNR分别提升89.5%和92.1%(P<0.05)。4组图像质量主观评分比较差异有统计学意义(P<0.05),从DLIR-H到50%ASiR-V依次降低。其中50%ASiR-V、DLIR-L组间比较差异没有统计学意义(P>0.05),其余各组间比较差异均有统计学意义(P<0.05)。各组间直肠癌局部侵犯情况诊断信心评分比较差异有统计学意义(P<0.05),DLIR-M及DLIR-H组优于50%ASiR-V组(P<0.05)。结论:与标准50%ASiR-V图像相比,DLIR-M和DLIR-H重建算法能有效提高图像质量,重建强度越高,图像质量越好,显示细微结构的能力越强,能为临床精准评估及个体化精准治疗提供更多的依据。Objective To improve the CT image quality of the anorectal junction in venous phase using a new deep learning image reconstruction(DLIR)algorithm.Methods A retrospective analysis was conducted on 71 patients undergoing pelvic computed tomography(CT)scans.All CT images were reconstructed at a thin slice thickness of 0.625 mm using 50%ASiR-V,low-,medium-and high-intensity DLIR(DLIR-L,DLIR-M and DLIR-H).The CT attenuations and standard deviation values of anal canal and hip fat were measured for each reconstruction group.With the standard deviation of hip fat as background noise,the contrast-to-noise ratio(CNR)and signal-to-noise ratio(SNR)of anal canal were calculated.Two radiologists independently assessed image quality and diagnostic confidence for local invasion of rectal cancer using the 5-point Likert scale.The objective measurement indicators and subjective scores were analyzed and compared,and Kappa test was used to evaluate the consistency.Results The differences in CT value of anal canal and hip fat among the groups were trivial(P>0.05),but fat SD,anal canal SNR and CNR(P<0.05)differed significantly,with lowest fat SD,highest anal canal SNR and CNR in DLIR-H group,while highest fat SD,lowest anal canal SNR and CNR in 50%ASiR-V group.Compared with 50%ASiR-V group,DLIR-H group decreased fat SD by 44.3%,but increased anal canal SNR and CNR by 89.5%and 92.1%,respectively(P<0.05).The subjective score of 4 groups were significantly different(P<0.05),decreasing from DLIR-H to 50%ASiR-V,and the inter-group differences were significant(P<0.05),except the difference between 50%ASiR-V group and DLIR-L group(P>0.05).There was a statistically significant difference in the diagnostic confidence for local invasion of rectal cancer among different groups(P<0.05),and the scores were significantly higher in DLIR-M and DLIR-H groups than in 50%ASiR-V and DLIR-L groups(P<0.05).Conclusion Compared with the standard 50%ASiR-V image,DLIR-M and DLIR-H reconstruction algorithms can effectively improve the image quality for the ano

关 键 词：直肠 电子计算机断层扫描 深度学习图像重建 图像质量 

分 类 号：R318[医药卫生—生物医学工程] R811.1[医药卫生—基础医学]