深度学习重建在低辐射剂量CT肺动脉成像中的临床应用价值

The clinical value of deep learning reconstruction in low-dose CT pulmonary angiography

目的探讨深度学习重建(DLR)较混合迭代重建(HIR)在降低CT肺动脉成像(CTPA)辐射剂量中的作用及对图像质量的影响。方法前瞻性纳入2020年12月至2021年4月在北京协和医院临床疑诊为急性肺动脉栓塞(APE)或因其他肺动脉疾病需行CTPA检查的患者100例,根据区组随机化分为HIR组、DLR组,每组50例,记录患者的性别、年龄及体质指数(BMI)。HIR组、DLR组噪声指数(SD)分别设置为8.8、15,其他扫描参数及对比剂注射方案相同,分别采用HIR、DLR算法重建。计算有效剂量(ED)及体型特异性扫描剂量(SSDE)。在1~3级肺动脉管腔、双侧椎旁肌勾画感兴趣区(ROI),记录各ROI CT值及标准差值,计算图像的信噪比(SNR)、对比噪声比(CNR)。由2名高年资医师采用盲法,以Likert 5分制对2组图像进行主观评估,评分不一致时由第3名医师综合判定。采用独立样本t检验对比2组患者一般资料、辐射剂量及客观图像质量,采用Mann-Whitney U检验对主观噪声、肺动脉分支显示、诊断信心进行组间比较,计算线性加权Kappa系数分析2名医师评分的一致性。结果HIR组、DLR组患者性别、年龄、BMI差异无统计学意义(P>0.05)。HIR组与DLR组图像1~3级肺动脉及椎旁肌CT值差异无统计学意义(P>0.05)。DLR组平均ED为1.3 mSv,SSDE中位数为4.20 mGy,均较HIR组降低约35%,但DLR组CTPA图像的SNR(30±5)、CNR(26±5)均高于HIR组(分别为23±5、20±5,t值分别为-6.60、-5.90,P<0.001)。DLR组主观图像噪声得分明显高于HIR组(Z=-7.34,P<0.001),且2名医师一致性高(Kappa=0.847,95%CI 0.553~1.000);DLR组与HIR组图像在肺动脉分支显示以及医师诊断信心得分上差异无统计学意义(Z分别为-0.259、-0.296,P>0.05)。结论DLR能够进一步降低CTPA检查辐射剂量,提高图像质量且不影响医师诊断信心,值得在临床推广。

Objective To explore the effect of deep learning reconstruction(DLR)on radiation dosage reduction and image quality of CTPA compared with hybrid iterative reconstruction(HIR).Methods A total of 100 patients with suspected pulmonary embolism(APE)or indications for CTPA due to other pulmonary artery diseases in Peking Union Medical College Hospital from December 2020 to April 2021 were prospectively enrolled and divided into HIR group and DLR group according to block randomization,with 50 cases in each group.The patient′s gender,age and body mass index(BMI)were recorded.HIR group and DLR group underwent standard deviation(SD)=8.8 and SD=15 CTPA protocols in combination with HIR and DLR algorithm respectively.Other scanning parameters and contrast medium injection plan were the same.The effective dose(ED)and size-specific dose estimate(SSDE)were calculated.Regions of interest(ROIs)were drawn in the lumen of Grade 1-3 pulmonary arteries and bilateral paravertebral muscles.The corresponding CT and SD values were recorded to acquire signal to noise ratio(SNR)and contrast noise ratio(CNR).Based on a double-blind method,two radiologists evaluated the subjective noise,visualization of pulmonary arteries,and diagnostic confidence of the two groups by 5-point Likert scales.The inconsistent results were judged comprehensively by the third radiologist.Independent samples t-test was used to compare the demographic data,radiation dosage and quantitative image quality of the two groups.Mann-Whitney U test was used to compare the subjective noise,visualization of pulmonary arteries and diagnostic confidence between the two groups.Linear weighted Kappa coefficient was calculated to analyze the consistency of the qualitative scores between the two radiologists.Results There were no significant differences in gender,age and BMI between the two groups(P>0.05).The CT values of Grade1-3 pulmonary arteries and paravertebral muscle had no significant differences(P>0.05).Compared with HIR group,the ED and SSDE in DLR group decreased by about 35%to 1.3 mSv and 4.20 mGy respectively,while the SNR(30±5)and CNR(26±5)of CTPA images were higher in DLR group than those in HIR group(23±5 and 20±5,with t=-6.60 and-5.90,respectively,both P<0.001).The subjective noise score was higher in DLR group than that in HIR group(Z=-7.34,P<0.001).In addition,two radiologists showed excellent interobserver agreement in DLR group(Kappa=0.847,95%CI 0.553-1.000).No significant differences were found in visualization of pulmonary arteries and diagnostic confidence between the two groups(P>0.05).Conclusion DLR further reduced the radiation dosage and improved the image quality of CTPA,with no detriment to diagnostic confidence.Thus DLR is worthy of clinical promotion.