文献速递胸部CT

深度学习图像重建在胸部CT成像中的临床价值:一项系统综述

Clinical value of deep learning image reconstruction in chest computed tomography (CT) imaging: a systematic review.
PubMed 文献 · Clin Radiol · 2025-11-21

本系统综述纳入13项比较深度学习图像重建(DLIR)与迭代重建(IR)/滤波反投影(FBP)的胸部CT研究,结果显示DLIR在结节检测敏感度(最高96.9%)、肺纹理分析AUC(0.97-1.0 vs 混合IR 0.91-0.97)及间质性肺病模式分类观察者间一致性(κ最高0.992)方面表现更优,并实现最高97%的剂量降低和更快的重建时间,结论为DLIR诊断性能不劣于甚至优于FBP/IR。

要点

  • DLIR在肺结节检测中敏感度最高达96.9%,优于IR/FBP。
  • DLIR肺纹理分析AUC为0.97-1.0,混合IR为0.91-0.97。
  • DLIR用于ILD模式分类的观察者间一致性κ最高达0.992。

核心结论

  • DLIR与FBP/IR相比诊断性能不劣于甚至更优,支持其在常规胸部CT中的应用;仍需大规模研究确立其对患者结局的影响并指导临床采纳。

研究设计与样本

  • 系统综述,遵循PRISMA指南并在PROSPERO注册,检索PubMed、Embase、Scopus、Web of Science、IEEE和Cochrane Library,纳入比较DLIR与IR/FBP的胸部CT人体研究,最终纳入13项研究,采用QUADAS-2进行质量评价,因结局异质性采用效应方向图和符号检验进行定性综合。

诊断标准与关键数值

  • DLIR结节检测敏感度最高达96.9%。
  • DLIR肺纹理分析AUC为0.97-1.0,混合IR为0.91-0.97。
  • DLIR用于ILD模式分类的观察者间一致性κ最高达0.992。
  • DLIR实现最高97%的剂量降低。

临床应用与阅片要点

  • DLIR在维持诊断一致性的同时可实现最高97%的剂量降低并缩短重建时间。
  • DLIR在结节检测、肺纹理分析和ILD模式分类方面诊断性能不劣于甚至优于IR/FBP。
  • 原文结论认为仍需大规模研究确立DLIR对患者结局的影响并指导临床采纳。

局限性

  • 因结局异质性,结果采用效应方向图和符号检验进行定性综合而非定量合并。
  • 原文结论指出仍需大规模研究确立DLIR对患者结局的影响并指导临床采纳。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

AIM: Computed tomography (CT) plays a central role in thoracic imaging, but maintaining diagnostic image quality at reduced doses remains a challenge. Filtered back projection (FBP) produces high noise, and iterative reconstruction (IR) reduces noise but alters image texture at low dose. Deep learning image reconstruction (DLIR) suppresses noise while preserving detail, yet its diagnostic performance in chest CT remains unclear. This review aimed to evaluate the clinical diagnostic value of DLIR in chest CT imaging.

MATERIALS AND METHODS: A systematic review was conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (International Prospective Register of Systematic Reviews [PROSPERO] registered). The following databases were searched for studies comparing DLIR with IR/FBP in chest CT: PubMed, Embase, Scopus, Web of Science, IEEE, and Cochrane Library. Eligible studies included human participants and reported diagnostic or image-quality outcomes. Quality assessment was performed using the QUADAS-2 tool. Given outcome heterogeneity, results were synthesised qualitatively using effect direction plots and sign tests.

RESULTS: From 1,967 records, 13 studies met the inclusion criteria. DLIR demonstrated superior diagnostic performance compared with IR/FBP and showed higher sensitivity for nodule detection (up to 96.9%), improved area under the curve (AUC) for lung texture analysis (0.97-1.0 vs 0.91-0.97 with hybrid IR), and stronger interobserver agreement for interstitial lung disease (ILD) pattern classification (κ up to 0.992). DLIR achieved substantial dose reductions (up to 97%) and faster reconstruction times while maintaining diagnostic consistency.

CONCLUSION: DLIR demonstrates noninferior to superior diagnostic performance compared with FBP/IR, supporting its role in routine chest CT. Large-scale studies remain essential to establish its impact on patient outcomes and guide clinical adoption.

来源:Clin Radiol,查看原文。版权归原作者所有。

引用

  1. Obhuli CM, Pendem S, Abhijith S, et al. Clinical value of deep learning image reconstruction in chest computed tomography (CT) imaging: a systematic review. Clin Radiol. 2025. PMID: 41411962. doi:10.1016/j.crad.2025.107184

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。