文献速递中枢神经肿瘤放疗放疗物理与技术

基于CNN的深度学习模型在胶质母细胞瘤放疗自动勾画中的评价:综述

Evaluation of CNN-based deep learning models for auto-contouring in glioblastoma radiotherapy: a review.
PubMed 文献 · Radiat Oncol · 2025-11-13

本综述检索截至2025年1月的文献,评价CNN、U-Net、nnU-Net和PKMI-Net等深度学习模型在胶质母细胞瘤放疗靶区自动分割中的应用。综述比较了各模型架构的特点、挑战与局限,指出nnU-Net因自配置能力和对不同影像模态的适应性而表现出最强的分割准确性。

要点

  • nnU-Net在分析的模型中分割准确性最强。
  • 深度学习自动勾画可提高准确性并减少观察者间差异。
  • 精确的肿瘤体积勾画对胶质母细胞瘤放疗计划至关重要。

核心结论

  • nnU-Net架构在分析的模型中表现最强,因其自配置能力和对不同影像模态的适应性而提供更优的分割准确性。

研究设计与样本

  • 本综述检索Google Scholar、PubMed、Scopus和Web of Science截至2025年1月的近期文献,评价CNN、U-Net、nnU-Net和PKMI-Net等深度学习模型在胶质母细胞瘤放疗自动勾画中的应用。

主要发现

  • 深度学习技术特别是CNN架构被引入自动勾画方法,以提高准确性并减少观察者间差异。
  • 综述比较了近期深度学习模型在胶质母细胞瘤靶区自动分割中的进展,聚焦更适用的架构及其特征、挑战和局限性。

临床怎么用

  • 放射治疗科可关注nnU-Net等深度学习模型在胶质母细胞瘤靶区自动勾画中的应用,以辅助提高勾画准确性和减少观察者间差异。

局限性

  • 综述聚焦于CNN、U-Net、nnU-Net和PKMI-Net等架构,未涉及其他类型模型。

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)。

原文

Glioblastoma (GBM) is the most common and aggressive primary brain tumor. Radiation therapy (RT) plays a main role in GBM treatment, with precise tumor volume delineation. Tumor contouring is important in RT treatment planning, followed by successful treatment. Recent advances in artificial intelligence (AI), particularly deep learning (DL) techniques, have introduced automated contouring methods to improve accuracy and reduce interobserver variability. This review considers different architectures such as CNN, U-Net, nnU-Net, and PKMI-Net. In addition, a comparison was made between recent advances in DL models for the automatic segmentation of GBM target volumes, focusing on more applicable architectures and their characteristics, challenges, and limitations. This review study was performed on the recent literature from scientific citation websites such as Google Scholar, PubMed, Scopus, and Web of Science by the end of January 2025. The results of this review study among the analyzed models, nnU-Net architecture emerged as the strongest, offering superior segmentation accuracy due to its self-configuring capabilities and adaptability to different imaging modalities. Not applicable.

来源:Radiat Oncol,查看原文。版权归原作者所有。

引用

  1. Soltaninejad A, Shahbazi-Gahrouei D, Khorasani A, et al. Evaluation of CNN-based deep learning models for auto-contouring in glioblastoma radiotherapy: a review. Radiat Oncol. 2025. PMID: 41233797. doi:10.1186/s13014-025-02748-y

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。