胸部低剂量CT联合深度学习重建算法在小儿支原体肺炎诊断中的可行性研究

Study on the feasibility of low-dose chest CT combined with deep learning reconstruction algorithm in the diagnosis of pediatric mycoplasma pneumonia

目的:探讨胸部低剂量(60kV)CT扫描技术联合深度学习重建算法(ClearInfinity)在小儿支原体肺炎诊断中的可行性。方法:选取2023年12月至2024年1月湖北医药学院附属太和医院收治的132例经临床确诊的支原体肺炎患儿,均行胸部CT扫描,并将其按照就诊序号随机分为常规剂量组(66例)和低剂量ClearView和ClearInfinity组(66例)。常规剂量组胸部CT扫描管电压100kV,采用50%ClearView迭代算法重建;低剂量ClearView和ClearInfinity组胸部CT扫描管电压均为60kV,分别使用50%ClearView迭代算法重建和50%ClearInfinity深度学习重建算法重建;比较3组辐射剂量的差异;测量并计算3组图像感兴趣区(ROI)的CT值和标准偏差值(SD),信噪比(SNR)和对比噪声比(CNR);由两名10年工作经验的影像诊断主治医师对图像进行主观评价,采用Kappa检验分析评分结果的一致性。结果:低剂量ClearView组和ClearInfinity组的容积CT剂量指数(CTDIvol)、剂量-长度乘积(DLP)、有效辐射剂量(ED)与常规剂量组比较,分别降低了87.58%、87.24%和88.00%,差异有统计学意义(t=4584.07、63.73、61.27,P<0.01)。常规剂量组的左、右肺噪声值低于低剂量ClearView组,而高于ClearInfinity组,差异有统计学意义(Z=-9.912、-7.013,P<0.01),低剂量ClearView组与ClearInfinity组比较,差异有统计学意义(Z=-9.912,P<0.01)。低剂量ClearView组的左、右肺SNR和CNR低于常规剂量组,差异有统计学意义(t=-34.810、5.522,P<0.01);低剂量ClearInfinity组的SNR和CNR高于常规剂量组,差异有统计学意义(t=3.544、-8.674,P<0.05)。两名主治医师对图像主观评价具有较好的一致性(Kappa>0.75,P<0.01);常规剂量组的主观评分与低剂量ClearInfinity组比较,差异无统计学意义(P>0.05),但优于低剂量ClearView组,差异有统计学意义(Z=-6.425,P<0.01)。结论:针对小儿支原体肺炎的患儿,60kV胸部低剂量CT结合ClearInfinity深度学习重建算法能在降低辐射的前提下保证图像质量,保障了诊断效果。

Objective:To explore the feasibility of 60 kV low-dose scanning technique on chest combined with ClearInfinity deep learning reconstruction algorithm in the diagnosis of pediatric mycoplasma pneumonia.Methods:A total of 132 pediatric patients,who admitted to Taihe Hospital Affiliated to Hubei Medical College and were diagnosed as mycoplasma pneumonia,were selected,and all of them underwent computed tomography(CT)scans on chest.They were randomly divided into routine dose group(66 cases),low dose ClearView and ClearInfinity group(66 cases).In the routine dose group,the tube voltage of CT scan on chest was 100 kV,and 50%ClearView iterative algorithm was adopted in this group.The tube voltage of CT scan on chest was 60kV in low dose ClearView and ClearInfinity group,and 50%ClearView iterative algorithm and 50%ClearInfinity deep learning reconstruction algorithm were used respectively to conduct reconstruction.The difference of radiation dose among the three groups was compared.The CT values and standard deviation(SD)values,signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)of the region of interest(ROI)of the images of 3 groups were measured and calculated.The images were subjectively evaluated by two diagnostic attending physicians with 10 years of work experience,and the Kappa test was adopted to analyze the consistency of the score results.Results:Compared with the routine dose group,the volume CT dose index(CTDIvol)values,dose-length product(DLP)values and effective radiation dose(ED)values of the low dose ClearView group and ClearInfinity group reduced respectively by 87.58%,87.24%and 88.00%,and the differences were statistically significant(t=4584.07,63.73,61.27,P<0.01).The noise values of left and right lung of the routine dose group were significantly lower than those of the low dose ClearView group,while were significantly higher than those of ClearInfinity group,and the differences were significant(Z=-9.912,-7.013,P<0.01),and the difference of them between low dose ClearView group and ClearInfinity group was significant(Z=-9.912,P<0.01).The SNR and CNR of left and right lung of low dose ClearView group were significantly lower than those of the routine dose group,with statistically significant(t=-34.810,5.522,P<0.01),while these of the low dose ClearInfinity group were significant higher than them of the routine dose group(t=3.544,-8.674,P<0.05),respectively.The two attending physicians had favorable consistency in the subjective evaluation for images(Kappa>0.75,P<0.01).The subjective score of the routine dose group was not significantly different with that of the low dose ClearInfinity group(P>0.05),but was significantly better than that of the low dose ClearView group(Z=-6.425,P<0.01).Conclusion:For pediatric patients with mycoplasma pneumonia,the 60 kV low dose CT on chest combined with ClearInfinity deep learning reconstruction algorithm can ensure image quality on the premise of reducing radiation,which can ensure the diagnostic effect.