双神经网络深度学习算法对不同低剂量上腹部CT图像质量的评估

Evaluation of upper abdominal CT image quality at various low radiation doses using dual-neural-network deep learning algorithm

摘要

目的：探讨国产宽体超高分辨率CT结合双神经网络深度学习重建技术（DLIR-CI）对上腹部不同辐射剂量方案CT图像质量的优化作用。方法：纳入33例行上腹部CT平扫的患者，采用2025年3月上市的320排宽体探测器CT进行扫描，根据扫描剂量的不同分为常规剂量（RD）组（120 kV, 350 mA）、低剂量1（LD1）组（120 kV, 175 mA）和低剂量2（LD2）组（120 kV, 70 mA）。所有RD组采用迭代重建算法（CV40%）重建，LD1组和LD2组均采用CV40%和深度学习重建算法（CI）（强度系数为20%/40%/60%/80%）重建，总共对99层图像进行质量评估，测量不同算法肝脏和肾脏的CT值及标准差（SD），并计算信噪比（SNR）。同时采用5分评价法评价所有图像的噪声等级和锐利度，客观评价采用SPSS软件的线性混合模型分析，事后两两比较采用Bonferroni法校正。结果：（1）与RD组的CTDIvol相比，LD1组和LD2组的剂量分别降低50%和80%；（2）与RD_CV40%相比，LD1_CV40%和LD2_CV40%的SD值分别增加24%和71%，SNR值分别降低21%和43%；（3）在LD1组和LD2组的组内比较中，CI40%、CI60%、CI80%的SD值和SNR值分别与CV40%相比，肝脏和肾脏的SD值均逐渐降低，SNR值均显著提高（P<0.05）；（4）LD1_CI40%与RD_CV40%相比，肝脏和肾脏的SD值和SNR值差异均无统计学意义（P>0.05）；（5）LD1组和LD2组的CI60%和CI80%分别与RD_CV40%相比，肝脏和肾脏的SD值更低，SNR值更高，差异均有统计学意义（P<0.05）；（6）主观评价中，LD1组和LD2组的CI40%、CI60%和CI80%的噪声等级评分和锐利度评分均明显优于RD_CV40%的评分，差异有统计学意义（P<0.05）。结论：上腹部CT扫描剂量较常规剂量降低50%、80%时，结合深度学习重建算法（强度系数≥40%）的图像可以达到常规剂量迭代重建CV40%的效果，甚至更优的图像质量，在临床应用中明显降低辐射风险。

Abstract

Objective To investigate the image quality optimization performance of a domestic wide-detector ultra-high-resolution CT combined with a dual-neural-network deep learning reconstruction technique (DLIR-CI) on upper abdominal CT scans acquired with different radiation dose protocols. Methods Thirty-three patients undergoing non-contrast upper abdominal CT scans were enrolled. All scans were obtained using a 320-slice wide-detector CT scanner released in March 2025. According to the scanning dose, patients were divided into a routine dose (RD) group (120 kV, 350 mA), a low dose 1 (LD1) group (120 kV, 175 mA), and a low dose 2 (LD2) group (120 kV, 70 mA). Iterative reconstruction algorithm (CV40%) was utilized in the RD group, while the LD1 and LD2 groups adopted CV40% and deep learning reconstruction algorithms (strength levels: 20%, 40%, 60%, and 80%). Quality evaluation was conducted on 99 image sets. The CT values and standard deviation (SD) of the liver and kidneys were measured, and the signal-to-noise ratio (SNR) was calculated. A 5-point scale was used to evaluate the noise level and sharpness of all images, and a linear mixed models analysis was conducted using SPSS software for objective assessment. Post hoc pairwise comparisons were corrected with the Bonferroni method. Results (1) Compared with the CTDIvol of the RD group, the radiation doses of LD1 and LD2 groups were reduced by 50% and 80%, respectively. (2) Compared with the RD_CV40%, the SD values of the LD1_CV40% and LD2_CV40% increased by 24% and 71%, while their SNR values decreased by 21% and 43%, respectively. (3) Intragroup comparisons of the LD1 and LD2 groups revealed that the SD values of the liver and kidneys gradually decreased and SNR values increased significantly at CI40%, CI60% and CI80% relative to CV40% (P<0.05). (4) No statistically significant differences were observed in the SD and SNR values of the liver and kidneys between LD1_CI40% and RD_CV40% (P>0.05). (5) Compared with RD_CV40%, the CI60% and CI80% in the LD1 and LD2 groups had lower SD values and higher SNR values in the liver and kidneys, with statistically significant differences (P<0.05). (6) Subjective evaluation showed that the noise and sharpness scores (score ≥3) of CI40%, CI60% and CI80% in the LD1 and LD2 groups were remarkably superior to those of CV40%, and the differences were statistically significant (P<0.05). Conclusion When the radiation dose for upper abdominal CT scanning is reduced by 50% and 80% relative to the conventional dose, the images reconstructed by deep learning reconstruction algorithm at strength levels ≥ 40% can achieve equivalent or even better image quality than conventional-dose images reconstructed by the RD_CV40% iterative algorithm. The proposed method can significantly reduce radiation exposure risks in clinical practice.

关键词

CT / 宽体探测器 / 深度学习重建算法 / 低剂量 / 图像质量

Key words

computed tomography / wide-detector / deep learning reconstruction algorithm / low-dose / image quality

参考文献

[1] Kwon H, Cho J, Oh J, et al. The adaptive statistical iterative reconstruction—V technique for radiation dose reduction in abdominal CT: comparison with the adaptive statistical iterative reconstruction technique[J]. Br J Radiol, 2015, 88(1054): 20150463.

[2] 浦仁旺, 刘义军, 刘静红, 等 . 低管电压结合自适应统计性迭代算法对腹部CT平扫图像的影响[J]. 中国临床医学影像杂志, 2015, 26(1): 66-68.

[3] Pu RW, Liu YJ, Liu JH, et al. Effects of low tube voltage combining with ASIR on abdominal plain CT scan images[J]. Journal of China Clinic Medical Imaging, 2015, 26(1): 66-68.

[4] 赵英明, 邓克学, 韦炜, 等 . 迭代算法对腹部CT图像影响及降低剂量的研究[J]. 中国医学计算机成像杂志, 2014, 20(5): 407-411.

[5] Zhao YM, Deng KX, Wei W, et al. Impact of iterative reconstruction algorithm on abdominal CT image quality and radiation dose[J]. Chinese Computed Medical Imaging, 2014, 20(5): 407-411.

[6] Christe A, Charimo—Torrente J, Roychoudhury K, et al. Accuracy of low—dose computed tomography (CT) for detecting and characterizing the most common CT—patterns of pulmonary disease[J]. Eur J Radiol, 2013, 82(3): e142-e150.

[7] Nelson RC, Feuerlein S, Boll DT . New iterative reconstruction techniques for cardiovascular computed tomography: how do they work, and what are the advantages and disadvantages?[J]. J Cardiovasc Comput Tomogr, 2011, 5(5): 286-292.

[8] Benz DC, Benetos G, Rampidis G, et al. Validation of deep—learning image reconstruction for coronary computed tomography angiography: impact on noise, image quality and diagnostic accuracy[J]. J Cardiovasc Comput Tomogr, 2020, 14(5): 444-451.

[9] Racine D, Becce F, Viry A, et al. Task—based characterization of a deep learning image reconstruction and comparison with filtered back—projection and a partial model—based iterative reconstruction in abdominal CT: a phantom study[J]. Phys Med, 2020, 76: 28-37.

[10] Kim I, Kang H, Yoon HJ, et al. Deep learning—based image reconstruction for brain CT: improved image quality compared with adaptive statistical iterative reconstruction—Veo (ASIR—V)[J]. Neuroradiology, 2021, 63(6): 905-912.

[11] McCollough CH, Bushberg JT, Fletcher JG, et al. Answers to common questions about the use and safety of CT scans[J]. Mayo Clin Proc, 2015, 90(10): 1380-1392.

[12] Marin D, Nelson RC, Rubin GD, et al. Body CT: technical advances for improving safety[J]. AJR Am J Roentgenol, 2011, 197(1): 33-41.

[13] Barca P, Giannelli M, Fantacci ME, et al. Computed tomography imaging with the adaptive statistical iterative reconstruction (ASIR) algorithm: dependence of image quality on the blending level of reconstruction[J]. Australas Phys Eng Sci Med, 2018, 41(2): 463-473.

[14] 魏巍, 杨旭, 童小雨, 等 . 深度学习重建算法联合低剂量增强CT对肝脏低对比度病灶显示的影响[J]. 中国医学计算机成像杂志, 2024, 30(3): 333-337.

[15] Wei W, Yang X, Tong XY, et al. Effect of deep learning reconstruction algorithm combined with low—dose contrast—enhanced CT on liver low—contrast lesion[J]. Chinese Computed Medical Imaging, 2024, 30(3): 333-337.

[16] Zeng LM, Xu X, Zeng W, et al. Deep learning trained algorithm maintains the quality of half—dose contrast—enhanced liver computed tomography images: comparison with hybrid iterative reconstruction: study for the application of deep learning noise reduction technology in low dose[J]. Eur J Radiol, 2021, 135: 109487.

[17] Nakamura Y, Higaki T, Tatsugami F, et al. Deep learning—based CT Image reconstruction: initial evaluation targeting hypovascular hepatic metastases[J]. Radiol Artif Intell, 2019, 1(6): e180011.

[18] 侯平, 刘杰, 陈岩, 等 . 基于深度学习的ClearInfinity算法在肝脏病变低剂量CT扫描中的应用价值[J]. 中国医疗设备, 2023, 38(5): 120-124.

[19] Hou P, Liu J, Chen Y, et al. Study on application value of ClearInfinity algorithm based on deep learning in low—dose CT scanning of liver lesions[J]. China Medical Devices, 2023, 38(5): 120-124.

[20] Neusoft Medical Systems Co., Ltd. Medical imaging using neural networks: US 11,250,543 B2[P]. 2022—02—15.

[21] 侯平, 刘杰, 陈岩, 等 . 基于深度学习的ClearInfinity算法对腹部体模CT扫描图像质量和辐射剂量的影响[J]. 中国医疗设备, 2022, 37(8): 99-103.

[22] Hou P, Liu J, Chen Y, et al. Impact of ClearInfinity algorithm based on deep learning on image quality and radiation dose of abdominal CT in phantom[J]. China Medical Devices, 2022, 37(8): 99-103.

[23] 毛玮斌, 孟占鳌, 覃杰, 等 . 深度学习图像重建联合器官剂量调制在颅脑低剂量CT中的图像质量与晶状体防护优化研究[J]. 新医学, 2025, 56(12): 1187-1197.

[24] Mao WB, Meng ZA, Qin J, et al. Deep learning image reconstruction combined with organ dose modulation for optimization in low—dose cranial CT: image quality and lens protection[J]. Journal of New Medicine, 2025, 56(12): 1187-1197.

[25] 徐灿飞, 黄艳, 张开治 . 机器学习结合磁共振弥散加权成像识别发病4.5 h内急性脑卒中患者[J]. 吉林大学学报(医学版), 2023, 49(6): 1635-1641.

[26] Xu CF, Huang Y, Zhang KZ . Acute ischemic stroke patients with onset time within 4.5 h identified by machine learning combined with magnetic resonance diffusion weighted imaging[J]. Journal of Jilin University (Medicine Edition), 2023, 49(6): 1635-1641.

基金资助

广东省自然科学基金(2024A1515012023)

南方医科大学南方医院院长基金(2024A009)