深度学习图像重建算法改善动态负荷心肌CT灌注成像图像质量的研究  

Study on deep learning image reconstruction to improve image quality in dynamic stress myocardial CT perfusion imaging

作　　者：区楚岚 曹励琪 虢梦雅 杨粤龙 杨峻青 刘畅[2] 陈佳玉 曹希明 李新云 刘辉 Ou Chulan;Cao Liqi;Guo Mengya;Yang Yuelong;Yang Junqing;Liu Chang;Chen Jiayu;Cao Ximing;Li Xinyun;Liu Hui(School of Medicine,South China University of Technology,Guangzhou 510006,China;Department of Radiology,Guangdong Provincial People's Hospital(Guangdong Academy of Medical Sciences),Southern Medical University,Guangzhou 510080,China;CT Imaging Research Center,GE Healthcare China,Beijing 100176,China;Department of Cardiology,Guangdong Provincial People's Hospital(Guangdong Academy of Medical Sciences),Southern Medical University,Guangzhou 510080,China)

机构地区：[1]华南理工大学医学院,广州510006 [2]南方医科大学附属广东省人民医院(广东省医学科学院)放射科,广州510080 [3]GE医疗中国CT影像研究中心,北京100176 [4]南方医科大学附属广东省人民医院(广东省医学科学院)心内科,广州510080

出　　处：《中华放射学杂志》2025年第1期27-35,共9页Chinese Journal of Radiology

基　　金：国家自然科学基金(82371903);广东省基础与应用基础研究基金(2024A1515012087)。

摘　　要：目的探讨深度学习图像重建算法(DLIR)相较于自适应统计迭代重建算法(ASiR-V)在改善动态负荷心肌CT灌注成像(CTP)图像质量及提高心肌边缘清晰度方面的能力。方法收集2023年9月至2024年2月在广东省人民医院行动态负荷心肌CTP的30例患者。对所有入组患者的影像资料分别使用ASiR-V 50%、ASiR-V 80%和中强度DLIR(DLIR-M)、高强度DLIR(DLIR-H)算法进行图像重建。在左心室腔、室间隔及左心室侧壁选取感兴趣区测量其CT值和标准差(SD),计算信噪比(SNR)和对比噪声比(CNR)。使用Matlab获得4个左心室心肌边缘CT值变化的差值(d)和CT值变化的斜率(s)用于评估客观边缘清晰度。由2名影像医师对图像的噪声、自然外观及边缘清晰度进行主观评分。2名医师评分不一致时由第3名高年资医师评分决定。分别计算SD值较低、SNR及CNR较高的ASiR-V和DLIR图像的左心室心肌血流量(MBF)。符合正态分布时,两组间比较采用独立样本t检验,多组间比较采用随机区组设计的方差分析;不符合正态性分布时,使用Friedman检验,两两比较采用Bonferroni校正检验。结果4种图像在室间隔、左心室侧壁的SD、SNR及CNR的差异均有统计学意义(P均<0.05),ASiR-V 80%与DLIR-H的SD值最低、SNR和CNR最高,主观图像噪声评分最高。4种图像在4个左心室心肌边缘的d和s总体差异均有统计学意义(P均<0.05),DLIR-M和DLIR-H的客观边缘清晰度最优[5(5,5)分],ASiR-V 80%最差[3.5(3,4)分]。4种图像的自然外观主观评分中,DLIR-M与DLIR-H评分最高,ASiR-V 80%最低[3(3,4)分],差异有统计学意义(P均<0.05)。使用ASiR-V 80%和DLIR-H图像分别计算的MBF值差异没有统计学意义(P均>0.05)。结论DLIR-H重建的动态负荷心肌CTP图像的SD、SNR及CNR与ASiR-V 80%相当,且采用DLIR-H可提高左心室心肌的边缘清晰度,同时不影响MBF值计算。ObjectiveTo explore the capability of deep learning image reconstruction(DLIR)compared to adaptive statistical iterative reconstruction(ASiR-V)in improving the image quality and myocardial edge sharpness of dynamic stress myocardial CT perfusion imaging(CTP).MethodsThirty subjects who underwent dynamic stress myocardial CTP at Guangdong Provincial People′s Hospital from September 2023 to February 2024 were recruited.Image data of all enrolled patients were reconstructed using ASiR-V 50%,ASiR-V 80%,medium-intensity DLIR(DLIR-M),and high-intensity DLIR(DLIR-H),respectively.Regions of interest were selected in the left ventricular cavity,interventricular septum,and left ventricular lateral wall for measurement of CT values and standard deviations(SD),and calculation of signal to noise ratio(SNR)and contrast to noise ratio(CNR).Matlab was utilized to obtain the differences(d)and slopes(s)of CT value changes at four left ventricular myocardial edges for objective edge sharpness evaluation.Two radiologists subjectively scored the images for noise,natural appearance,and edge sharpness.In case of disagreement between the two radiologists,a third senior radiologist′s score was decisive.Left ventricular myocardial blood flow(MBF)of ASiR-V and DLIR images with lower SD,higher SNR and CNR were calculated,respectively.When the normal distribution was satisfied,the independent sample t test was used for comparison between two groups,and the random block design ANOVA was used for comparison between multiple groups.And analysis was conducted using Friedman test for non-normally distributed data,and Bonferroni correction for pairwise comparisons.ResultsThere were statistically significant differences in SD,SNR,and CNR among the four images in the interventricular septum and left ventricular lateral wall(all P<0.05),with ASiR-V 80%and DLIR-H demonstrating the lowest SD,highest SNR and CNR,and the subjective image noise score.Statistically significant differences were observed in d and s for the four left ventricular myocardial

关 键 词：体层摄影术 X线计算机 心肌灌注成像 深度学习重建 图像质量 边缘清晰度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] R816.2[医药卫生—放射医学]