《中国医学装备》杂志社

目的：通过体模实验评估60 kV超低剂量CT扫描结合深度学习重建算法(CI算法)在儿童胸部中的可行性。方法：采用PH-1C型儿童胸部体模，根据照射剂量参数的不同，分为常规组(标准照射剂量：100 kV/60 mA)与实验组[60 kV/不同管电流(60、30、15、8、5 mA)]。常规组重建采用权重为50%的常规迭代算法(CV算法)；实验组重建分别采用50%CV算法、以及不同权重(10%、30%、50%、70%、90%)的CI算法。比较实验组内不同重建算法(50%CV、10%CI、30%CI、50%CI、70%CI、90%CI)的主观及客观图像差异；比较实验组[60 kV(60、30、15、8、5 mA)]与常规组之间的主观及客观图像差异和辐射剂量差异。结果：在实验组内，不同权重重建算法(50%CV、10%CI、30%CI、50%CI、70%CI、90%CI)的图像质量比较时，CI算法重建的图像质量显著优于CV算法重建的图像，差异有统计学意义(F=89.42、5.37,P<0.05)。其中50%CI、70%CI、90%CI客观评分高于10%CI和30%CI，但三者间的差异均无统计学意义(P>0.05)；主观评价中70%CI图像质量最佳。实验组[60 kV/不同管电流(60、30、15、8、5 mA)]均采用70%CI算法重建的图像与常规组图像比较，实验组中60 kV/60 mA的图像质量主观评价及客观评分与常规组比较，差异均无统计学意义(P>0.05)。60 kV/30 mA的图像质量肺窗客观评分与常规组比较，差异均无统计学意义(P>0.05)，主观评价略低于常规组；纵隔窗在客观评分中SD值(12.85±0.65)高于常规组，SNR值(5.07±0.32)低于常规组，差异有统计学意义(t=14.82,2.26,P<0.05)，主观评价低于常规组。实验组不同管电流时CT辐射剂量参数均显著低于常规组，差异均有统计学意义(t=215.90、220.01、231.35、237.72、249.70、344.57、368.65、379.50、387.64、392.08、267.48、298.72、301.12、308.07、319.80,P<0.05)。其中60 kV的60和30 mA组的有效剂量均值分别为0.11和0.05 mSv。结论：结合70%CI重建算法，60 kV/60 mA图像质量可达或略优于常规100 kV的水平；对于重点观察肺窗图像的患儿(肺炎复查者)，管电流可进一步降至30 mA。

Objective: This study aimed to assess the feasibility of 60 kV ultra-low-dose computed tomography(CT) scan combined with deep learning-based reconstruction algorithm for children's chest through a phantom experiment. Methods: This study adopted pediatric chest phantom of PH-1C type. Based on different exposure parameters, the scans were divided into a conventional group(standard dose: 100 kV/60 mA) and experimental group [60 kV/with different tube currents: 60, 30, 15, 8, 5 mA)]. The conventional group adopted conventional iterative algorithm(CV algorithm) that weight was 50% to conduct reconstruction, while experimental group adopted respectively 50% CV algorithm and CI algorithm based on deep-learning with different weight(10%, 30%, 50%, 70%, 90%) to conduct reconstruction. The differences of subjective and objective images among different algorithms(50%CV, 10%CI, 30%CI, 50%CI, 70%CI and 90%CI) within experimental group were compared. The differences of subjective and objective images, and exposure doses were compared between the experimental group [60 kV(60, 30, 15, 8, 5 mA) ] and the conventional group. Results: Within the experimental group, the reconstructed image quality of CI algorithm was significantly better than that of CV algorithm when the image qualities were compared among different weight reconstructed algorithm, and the differences were significant(F=89.42, 5.37, P<0.05). In them, the objective scores of 50%CI, 70%CI, and 90%CI were higher than those of 10%CI and 30%CI, but the differences among the three indicators were not statistically significant(P>0.05). In subjective evaluation, the image quality of 70%CI was the best. The experimental group [60 kV/different tube current(60, 30, 15, 8, 5 mA)] adopted the image that was reconstructed by 70% CI to compare with the image of conventional group, and there were not significant difference at subjective and objective evaluation of image quality between 60 kV/60 mA subgroup of experimental group and conventional group(P>0.05). There was not significant difference at objective score of lung window between 60 kV/30 mA subgroup and conventional group(P>0.05), and the subjective score of lung window of 60 kV/30 mA subgroup was slightly less than that of conventional group. The standard deviation(SD) value(12.85±0.65) in objective scores of mediastinal window was higher than that of conventional group, and the signal noise rate(SNR) value(5.07±0.32) was lower than that of conventional group, and the differences were significant(t=14.82, 2.26, P<0.05), while the subjective evaluation was lower than that of conventional group. The parameters of CT radiation dose of all subgroups in the experimental group were significantly lower than those in the conventional group, and the differences were statistically significant(t=215.90, 220.01, 231.35, 237.72, 249.70, 344.57, 368.65, 379.50, 387.64, 392.08, 267.48, 298.72, 301.12, 308.07, 319.80, P<0.05). In them, the mean values of effective dose of 60 and 30 mA group of 60 kV were respectively 0.11 and 0.05 m Sv. Conclusion: Combined with the 70%CI reconstruction algorithm, the image quality of 60 kV/60 mA can reach or slightly better to the level of conventional 100 kV. For pediatric patients(who receive re-examination for pneumonia) who were especially observed for the images of lung window, and the tube current can be further reduced to 30 mA.