Page 19 - 《中国医疗器械杂志》2026年第1期

Chinese Journal of Medical Instrumentation 2026年 第50卷 第1期

医 学 人 工 智 能

文章编号：1671-7104(2026)01-0015-09

深度学习下儿童髓母细胞瘤放疗脑及椎骨

亚结构自动分割算法研究

【作 者】 王为，陈淑贤，蒋马伟
上海交通大学医学院附属新华医院 肿瘤科，上海市，200092
【摘 要】 为评估nnU-Net和FuseNet模型在儿童髓母细胞瘤亚结构自动分割中的应用可行性，该文回顾性分析60例接
受放疗的患儿，以5岁为界分为两组（≤5岁组和＞5岁组），基于CT-MRI融合图像勾画脑亚结构，基于
CT图像勾画椎骨亚结构，训练U-Net、nnU-Net和FuseNet 3种卷积神经网络模型并评估结果，每组设训练
集24例，测试与验证集6例，另经20例外部独立队列验证泛化性。比较3种模型与图谱库（Atlas）法的
DSC，评估nnU-Net和FuseNet的HD95、RAVD等几何指标及人工修正耗时。结果显示，FuseNet在脑亚
结构分割中表现最优，在两组椎骨亚结构分割上均优于Atlas、U-Net（P=0.028、0.005 和P=0.005、
0.005），与nnU-Net无显著差异（P=0.107、0.236）。在≤5岁组中，FuseNet除小脑前叶和海马外，在
>5岁组中除海马外，其余亚结构DSC均值均>0.8，且两年龄组人工修正耗时均最短。结论表明，nnU-
Net可实现较好分割，FuseNet通过多模态特征动态融合提升脑亚结构分割精度，且修正效率最高。
【关 键 词】 自动分割；多模态卷积神经网络；儿童放疗；髓母细胞瘤；亚结构
【中图分类号】 R815; R318
【文献标志码】 A doi: 10.12455/j.issn.1671-7104.250219
Deep Learning-Based Automated Segmentation Algorithms of Brain and
Vertebral Substructures for Radiotherapy in Pediatric Medulloblastoma

【 Authors 】 WANG Wei, CHEN Shuxian, JIANG Mawei
Oncology Department, Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine,
Shanghai, 200092
【 Abstract 】 To evaluate the application feasibility of nnU-Net and FuseNet for automatic segmentation of pediatric
medulloblastoma substructures, 60 pediatric patients who received radiotherapy and grouped by 5-year
age (≤5 and >5 years) were retrospectively analyzed. Brain substructures were delineated on CT-MRI
fusion images, and vertebral substructures were delineated on CT images. The three convolutional neural
network models of U-Net, nnU-Net and FuseNet were trained (24 cases/group) and tested/validated
(6 cases/group), with 20 external cases verifying generalization. The three models and Atlas-based
methods were compared using DSC; nnU-Net/FuseNet’s HD95, RAVD and manual correction time were
evaluated. Results showed that FuseNet performed best in brain segmentation, outperforming Atlas and
U-Net in vertebral substructures for both age groups (P=0.028/0.005 and 0.005/0.005) but not differing
from nnU-Net (P=0.107/0.236). Its DSC exceeded 0.8 for most substructures (except cerebellar anterior
lobe/hippocampus in group≤5 years; hippocampus in group>5 years), with shortest correction time in
both age groups. This study has demonstrated that nnU-Net can achieve good segmentation; FuseNet
can improve brain segmentation accuracy via dynamic multimodal feature fusion, with the highest
correction efficiency.
【Key words】 automated segmentation, multimodal convolutional neural network, pediatric radiotherapy,
medulloblastoma, substructure

收稿日期：2025-04-01 0 引言
基金项目：2023年上海市中西医结合学会医学工程专业委员会课
题研究专项基金项目（YG202311）
作者简介：王为，E-mail: iewiew18@126.com 髓母细胞瘤（medulloblastoma, MB）是儿童最
通信作者：蒋马伟，E-mail: jiangmawei@xinhuamed.com.cn 常见的颅内恶性肿瘤，在5~10岁儿童中较常见，

15