实用医学杂志 ›› 2025, Vol. 41 ›› Issue (21): 3428-3434.doi: 10.3969/j.issn.1006-5725.2025.21.019

• 医学检查与临床诊断 • 上一篇    

深度学习重建算法联合轴扫的低剂量胸部CT在儿童肺炎支原体肺炎中的诊断价值

韩林梅,任盈丽,李依蔓,黄芬,杜涛明()   

  1. 成都市第七人民医院放射科 (四川 成都 610000 )
  • 收稿日期:2025-08-08 出版日期:2025-11-10 发布日期:2025-11-13
  • 通讯作者: 杜涛明 E-mail:dtm0528@sina.com
  • 基金资助:
    四川省医学会肿瘤/麻醉/放射/慢病/神经病(恒瑞-一行)专项科研项目(2024HR55)

Feasibility study of low⁃dose chest CT with deep learning reconstruction algorithm combined with axial scan in children with mycoplasma pneumoniae pneumonia

Linmei HAN,Yingli REN,Yiman LI,Fen HUANG,Taoming. DU()   

  1. Department of Radiology,NO. 7 people′s Hospital of Chengdu,Chengdu 610000,Sichuan,China
  • Received:2025-08-08 Online:2025-11-10 Published:2025-11-13
  • Contact: Taoming. DU E-mail:dtm0528@sina.com

摘要:

目的 探索深度学习重建算法联合轴扫的低剂量胸部计算机断层扫描(CT)在儿童肺炎支原体肺炎(MPP)中的诊断价值,以期为临床工作提供参考。 方法 选取2024年2月至2025年6月儿童MPP 160例作为研究对象,均行胸部CT检查,扫描方案为低剂量轴扫,分别利用深度学习图像重建(DLIR)算法与常规自适应迭代重建(ASIR-V)进行图像重建,比较DLIR与ASIR-V的客观图像质量[背景噪声(SD)、信号噪声比(SNR)、对比噪声比(CNR)]、主观图像质量、CT征象检出率,并对比DLIR、ASIR-V诊断MPP严重程度与临床诊断的一致性。 结果 随着DLIR强度和ASIR权重增加,SD逐渐降低,SNR、CNR逐渐升高,且DLIR-H SD低于ASIR-V80%,SNR、CNR高于ASIR-V80%(P < 0.05);Ridit检验显示,不同DLIR强度下DLIR-H的主观图像质量评分最优,不同ASIR权重下ASIR-V80%的主观图像质量评分最优,且DLIR-H的主观图像质量评分优于ASIR-V80%,差异有统计学意义(P < 0.05);DLIR-H条件下空气支气管征、肺实变影、间质浸润检出率分别为69.38%、86.88%、20.63%,高于ASIR-V80%的50.00%、71.88%、7.50%,差异有统计学意义(P < 0.05);一致性分析结果显示,DLIR-H条件下诊断MPP严重程度结果与临床诊断的Kappa值为0.856(95%CI:0.711 ~ 0.996),ASIR-V80%条件下诊断结果与临床诊断的Kappa值为0.498(95%CI:0.346 ~ 0.650);ROC分析结果显示,DLIR-H、ASIR-V80%诊断MPP严重程度的曲线下面积(AUC)分别为0.925(95%CI:0.872 ~ 0.960)、0.729(95%CI:0.653 ~ 0.796),DLIR-H诊断价值优于ASIR-V80%(Z = 3.952,P < 0.001)。 结论 深度学习重建算法可有效提高图像质量,DLIR-H联合轴扫的低剂量胸部CT对MPP严重程度具有较高诊断价值,可作为临床诊断MPP严重程度及降低辐射剂量的可行性方案。

关键词: 重建算法, 胸部CT, 低剂量轴扫, 图像质量, 儿童, 肺炎支原体肺炎

Abstract:

Objective To explore the diagnostic value of deep learning image reconstruction (DLIR) combined with low?dose chest computed tomography (CT) with axial scan in the diagnosis of mycoplasma pneumoniae pneumonia (MPP) in children, and to provide reference for clinical practice. Methods 160 cases MPP children from February 2024 to June 2025 were selected as study subjects, and low?dose chest CT with axial scan was performed on all patients. DLIR and conventional adaptive iterative reconstruction?V (ASIR?V) were used for image reconstruction. The objective image quality [background noise (SD), signal?to?noise ratio (SNR), and contrast?to?noise ratio (CNR)], subjective image quality, and CT sign detection rate were compared, and the consistency of DLIR and ASIR?V in the diagnosis of MPP severity and clinical diagnosis was compared. Results As the intensity of DLIR and the weight of ASIR increasd, SD gradually decreased, while SNR and CNR gradually increased. The high?strength DLIR (DLIR?H) SD was lower than that of ASIR with a blending level of 80% (ASIR?V80%). The SNR and CNR were higher than those of ASIR?V80%, showing statistical significance (P < 0.05). Ridit test showed that DLIR?H had the best subjective image quality score under different DLIR intensities, and ASIR?V80% had the best subjective image quality score under different ASIR weights. Furthermore, the subjective image quality score of DLIR?H was higher that of ASIR?V80%, and the differences were statistically significant (P < 0.05). Using DLIR?H, the detection rates of air bronchogram, pulmonary consolidation, and interstitial infiltration (69.38%, 86.88%, 20.63%, respectively) were higher than those using ASIR?V80% (50.00%, 71.88%, 7.50%, respectively), and the differences were statistically significant (P < 0.05). Consistency analysis showed that the Kappa value between the diagnostic results of MPP severity using DLIR?H and clinical diagnosis was 0.856 (95%CI: 0.771 ~ 0.996), while that between the diagnostic results of MPP severity using ASIR?V80% and clinical diagnosis was 0.498(95%CI: 0.346 ~ 0.650). ROC analysis showed that the area under the curve (AUC) for diagnosing MPP severity was 0.925 (95%CI: 0.872 ~ 0.960) for DLIR?H and 0.729 (95%CI: 0.653 ~ 0.796) for ASIR?V80%, and the diagnostic value of DLIR?H was superior to that of ASIR?V80% (Z = 3.952, P < 0.001). Conclusion DLIR can effectively improve image quality. DLIR?H combined with low?dose chest CT with axial scan has high diagnostic value for the severity of MPP, and can serve as a feasible solution for clinical diagnosis of MPP severity and reducing radiation dose.

Key words: reconstruction algorithm, chest CT, low-dose axial scan, image quality, children, mycoplasma pneumoniae pneumonia

中图分类号: