应用深度学习图像重建算法提高低辐射剂量肝脏能谱CT图像质量和病灶诊断信心的临床研究  

A clinical study of deep learning image reconstruction algorithms in liver dual-energy CT with reduced radiation dose to further improve image quality and lesion diagnostic confidence

作　　者：李云成 李裕国 杨君琳 宋建[1] 唐星 邓炜 王震 杨金秀 刘斌[1] 余永强[1] 李小虎[1] Li Yuncheng;Li Yuguo;Yang Junlin;Song Jian;Tang Xing;Deng Wei;Wang Zhen;Yang Jinxiu;Liu Bin;Yu Yongqiang;Li Xiaohu(Departmeny of Radiology,the First Affiliated Hospital of Anhui Medical University,Hefei 230022,China)

机构地区：[1]安徽医科大学第一附属医院医学影像科,合肥230022

出　　处：《中华放射学杂志》2025年第1期43-49,共7页Chinese Journal of Radiology

基　　金：国家自然科学基金(82071897、81970446);安徽省自然科学基金(2308085Y48);安徽省高校杰出青年科研项目(2022AH020071);安徽医科大学研究生科研与实践创新项目(YJS20230126)。

摘　　要：目的探讨使用深度学习图像重建(DLIR)提高肝脏低辐射剂量能谱CT图像质量、病灶诊断信心及碘浓度(IC)测量精度的可行性。方法本研究为前瞻性队列研究。前瞻性收集2023年6月至2024年1月在安徽医科大学第一附属医院接受肝脏增强CT的60例患者,采用随机区组法分为常规辐射剂量组和低辐射剂量组,各30例。常规辐射剂量组于门静脉期采用常规辐射剂量扫描,管电压120 kVp;低辐射剂量组接受低辐射剂量扫描,即管电压80 kVp和140 kVp快速切换能谱扫描模式。计算2组患者的有效辐射剂量。常规辐射剂量组采用40%强度的自适应统计迭代重建算法(ASIR-V)重建图像(AR40120 kVp)。低辐射剂量组采用高强度DLIR(DLIR-H)重建40、50 keV虚拟单能量图像(VMI)(DH-VMI 40 keV、DH-VMI 50 keV)。对上述3组图像质量进行客观评价,测量图像噪声,计算肝脏和门静脉的对比噪声比(CNR)和信号噪声比(SNR);对图像质量进行主观评分,评分内容包括图像噪声、对比度、病变显示和诊断信心。在低辐射剂量组中,采用DLIR-H、ASIR-V40%重建碘图,测量肝脏和门静脉IC值、标准差(SD),并计算变异系数(CV)。采用单因素方差分析或Kruskal-Wallis H检验比较3组图像间主客观图像质量评价指标的差异,采用配对t检验比较DLIR-H、ASIR-V40%重建碘图间测量指标的差异。结果低辐射剂量组的有效辐射剂量[(2.2±0.5)mSv]较常规辐射剂量组[(5.4±1.4)mSv]降低56.8%。客观评价中,DH-VMI 40 keV的图像噪声、肝脏和门静脉CNR和SNR均高于AR40120 kVp(P<0.001);DH-VMI 50 keV的图像噪声低于AR40120 kVp(P=0.020),肝脏和门静脉CNR和SNR高于AR40120 kVp(P<0.001)。主观评价中,DH-VMI 40 keV与AR40120 kVp间图像噪声评分差异无统计学意义(P>0.05),DH-VMI 50 keV的图像噪声评分低于AR40120 kVp(P<0.05);DH-VMI 40 keV和DH-VMI 50 keV的对比度、病变显示和诊断信心评分均高于AR40120 kVp(P<0.05)。在低辐射剂量组�ObjectiveTo explore the feasibility of applying deep learning image reconstruction(DLIR)in low-radiation dose liver dual-energy CT to further improve image quality,diagnostic confidence of lesion,and accuracy of iodine concentration(IC)measurement.MethodsThis prospective cohort study enrolled 60 patients scheduled for enhanced liver CT at the First Affiliated Hospital of Anhui Medical University from June 2023 to January 2024.The participants were randomly assigned into the standard dose group and low radiation dose group with 30 cases in each using randomized block method.The standard radiation dose group underwent standard-radiation dose 120 kVp scans during the venous phase,while the low radiation dose group underwent low radiation dose scans with a rapid kVp-switching spectral scanning mode at 80 kVp and 140 kVp.The effective radiation dose(ED)was calculated for both groups.The standard radiation dose group was reconstructed using adaptive statistical iterative reconstruction-V(ASIR-V)algorithm 40%(AR40120 kVp).The low radiation dose group using high-intensity DLIR(DLIR-H)to reconstructed 40 keV and 50 keV virtual monoenergetic images(VMI)(DH-VMI 40 keV,DH-VMI 50 keV).The image quality of the above three groups was objectively evaluated through the measurement of image noise and calculation of contrast-to-noise ratio(CNR)and signal-to-noise ratio(SNR)for the liver and portal vein;and the image quality was subjectively scored for image noise,contrast,lesion conspicuity,and diagnostic confidence.In the low radiation dose group,DLIR-H and ASIR-V40%reconstructed iodine maps were used to measure the liver and portal vein of IC values,standard deviations(SD),and coefficients of variation(CV).One-way analysis of variance or Kruskal-Wallis H test was used to compare the differences of subjective and objective image quality among the three groups,and paired t-test was used to compare the differences in measurement indexes between DLIR-H and ASIR-V40%reconstructed iodine maps.ResultsThe ED in the low radiation dose gro

关 键 词：体层摄影术 X线计算机 深度学习 图像重建 辐射剂量 

分 类 号：R816.5[医药卫生—放射医学] R575[医药卫生—临床医学]