深度学习算法结合三低技术在上腹部动脉CT血管造影中的应用

Application of deep learning algorithm combined with"triple-low"technique in CT angiography of upper abdominal arteries

目的探讨深度学习图像重建(DLIR)结合低辐射剂量、低对比剂剂量、低对比剂注射速度(三低技术)在上腹部动脉CT血管造影(CTA)中的应用价值。方法本研究对象为2021年6月至2021年10月在中山大学附属第三医院接受上腹部动脉CTA检查的60例患者。患者均签署知情同意书,符合医学伦理学规定。其中男33例,女27例;年龄19~86岁,中位年龄49岁。按扫描方案分为标准方案组(S组,30例)和三低方案组(L组,30例)。记录两组的有效辐射剂量(ED)、对比剂剂量、对比剂注射速度。对S组进行60%ASIR-V(S-AV60)图像重建;对L组进行60%ASIR-V(L-AV60)、80%ASIR-V(L-AV80)、DLIR-M(L-DM)和DLIR-H(L-DH)图像重建。客观图像质量评价参数包括上腹部动脉CT值、标准差(SD)值、信噪比(SNR)和对比噪声比(CNR);主观图像质量由两名放射科医师对重建图像进行双盲法评分。两组ED、对比剂剂量等比较采用t检验;SD、SNR、CNR等客观图像质量评价参数比较采用单因素方差分析;主观图像质量评分比较采用Kruskal-Wallis检验;采用Kappa检验分析两名放射科医师主观评分的一致性。结果L组ED为(5.1±1.3)mSv,明显低于S组的(10.5±2.1)mSv(t=-12.397,P<0.05);对比剂剂量为(65±11)ml,明显低于S组的(100±21)ml(t=-8.150,P<0.05);与S组注射速度5.0 ml/s相比,L组3.5 ml/s降低30%。对于上腹部CTA成像,L-DH组SD最小,SNR和CNR最大(P<0.05)。L-DH组重建图像的清晰度、图像噪声、图像伪影、小分支显示、临床诊断5个参数的主观图像质量评分分别为(4.7±0.5)、(4.6±0.5)、(4.8±0.4)、(4.5±0.5)、(4.7±0.5)分。5组重建方式中,L-DH组的主观图像质量评分最高(H=118.424,114.258,113.367,121.463,118.778;P<0.05)。两名放射科医师对5组上腹部动脉CTA的主观图像质量评分有较好的一致性(κ=0.672,P<0.05)。结论在上腹部动脉CTA中,三低技术可显著降低辐射剂量、对比剂剂量和对比剂注射速度。与推荐的60%ASIR-V标准方案相比,高级别DLIR结合三低技术可进一步改善图像质量,是较好的重建算法。

Objective To evaluate the application value of deep learning image reconstruction(DLIR)combined with low radiation dose,low contrast agent dose and low contrast agent injection speed(triple-low technique)in the CT angiography(CTA)of upper abdominal arteries.Methods 60 patients receiving CTA of the upper abdominal arteries in the Third Affiliated Hospital of Sun Yat-sen University from June to October 2021 were recruited in this study.The informed consents of all patients were obtained and the local ethical committee approval was received.Among them,33 patients were male and 27 female,aged from 19 to 86 years,with a median age of 49 years.According to the scanning plan,all patients were divided into the standard plan group(group S,n=30)and triple-low plan group(group L,n=30).The effective dose(ED),contrast agent dose and contrast agent injection speed in two groups were recorded.60%ASIR-V(S-AV60)image reconstruction was performed in group S,and 60%ASIR-V(L-AV60),80%ASIR-V(L-AV80),DLIR-M(L-DM)and DLIR-H(L-DH)image reconstructions were conducted in group L.Parameters of objective evaluation of image quality included CT value,standard deviation(SD)value,signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)of the upper abdominal artery.For the subjective image quality evaluation,the reconstructed images were scored with double-blind manner by two radiologists.ED and contrast agent dose were compared between two groups by t test.SD,SNR,CNR and other objective image quality evaluation parameters were compared by one-way ANOVA.Subjective image quality score was compared by Kruskal-Wallis test.The consistency of subjective score between two radiologists was assessed by Kappa test.Results In group L,ED was(5.1±1.3)mSv,significantly lower than(10.5±2.1)mSv in group S(t=-12.397,P<0.05).The contrast agent dose in group L was(65±11)ml,significantly lower than(100±21)ml in group S(t=-8.150,P<0.05).Compared with the injection speed of 5.0 ml/s in group S,the injection speed was 3.5 ml/s in group L,which was decreased by 30%.For CTA image of the upper abdomen,the SD in L-DH group was the smallest,and the SNR and CNR were the largest(P<0.05).In L-DH group,the subjective image quality scores of five parameters including sharpness,image noise,image artifact,small branch display and clinical diagnosis were 4.7±0.5,4.6±0.5,4.8±0.4,4.5±0.5 and 4.7±0.5,respectively.Among the five reconstructions,the subjective image quality score in the L-DH group was the highest(H=118.424,114.258,113.367,121.463,118.778;P<0.05).Fair consistency of 5 subjective image quality scores of CTA of upper abdominal arteries by 2 radiologists were observed(κ=0.672,P<0.05).Conclusions In the CTA of upper abdominal arteries,"triple-low"technique can significantly reduce the radiation dose,contrast agent dose and contrast agent injection speed.Compared with the recommended 60%ASIR-V standard plan,high-level DLIR combined with"triple-low"technique can further improve the image quality,which is a favorable reconstruction algorithm.