[1]李国鹏,肖洪兵,赵晓睿.基于多尺度扩展注意力的特征调制脑肿瘤全自动分割[J].中国医学物理学杂志,2026,43(1):76-83.[doi:DOI:10.3969/j.issn.1005-202X.2026.01.010]
　LI Guopeng,XIAO Hongbing,ZHAO Xiaorui.Fully automated brain tumor segmentation based on multi-scale expanded attention and feature modulation[J].Chinese Journal of Medical Physics,2026,43(1):76-83.[doi:DOI:10.3969/j.issn.1005-202X.2026.01.010]
点击复制

基于多尺度扩展注意力的特征调制脑肿瘤全自动分割()

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
43卷
期数:
2026年第1期
页码:
76-83
栏目:
医学影像物理
出版日期:
2026-01-26

文章信息/Info

Title:
Fully automated brain tumor segmentation based on multi-scale expanded attention and feature modulation
文章编号:
1005-202X（2026）01-0076-08
作者:
李国鹏; 肖洪兵; 赵晓睿
北京工商大学计算机与人工智能学院， 北京 100048
Author(s):
LI Guopeng;  XIAO Hongbing;  ZHAO Xiaorui
School of Computer and Artificial Intelligence, Beijing Technology and Business University, Beijing 100048, China
关键词:
脑肿瘤分割; 多尺度扩展注意力; 空间选择交互调制; 深度学习
Keywords:
Keywords: brain tumor segmentation multi-scale expanded attention spatially choice and cross modulation deep learning
分类号:
R318；TP181
DOI:
DOI:10.3969/j.issn.1005-202X.2026.01.010
文献标志码:
A
摘要:
针对脑肿瘤MRI影像分割中肿瘤形态复杂、边界模糊的问题，提出一种基于多尺度扩展注意力（MSEA）与空间选择交互调制（SCCM）的三维分割框架（MSEFA?et）。MSEA模块通过多膨胀率注意力头扩大感受野，增强对肿瘤复杂结构的理解；其稀疏自注意力操作在局部窗口内采样键值，在建模长程依赖的同时保留细节。SCCM机制通过特征融合与动态调制，自适应强化相关特征并抑制背景。在BraTS2021数据集上的实验表明，MSEFA2Net对肿瘤完整区域、肿瘤核心区域和肿瘤增强区域的Dice系数分别达到91.13%、87.46%和87.98%，验证其高效性与准确性。
Abstract:
Abstract: To address the issues of complex tumor morphology and blurred boundaries in brain tumor MRI image segmentation, a novel three-dimensional segmentation framework named MSEFA?et which integrates multi-scale expanded attention and spatially choice and cross modulation is proposed. Specifically, multi-scale expanded attention module leverages attention heads with various dilation rates to expand the receptive field, thereby enhancing the understanding of complex tumor structures, and its sparse self-attention samples key values within local windows for preserving detailed information while modeling long-range dependencies. Spatially choice and cross modulation mechanism adaptively strengthens relevant features and suppresses background interference through feature fusion and dynamic modulation. Experiments on the BraTS2021 dataset show that MSEFA?et achieves Dice scores of 91.13%, 87.46%, and 87.98% for the whole tumor, tumor core, and enhancing tumor, respectively, validating its superior efficiency and accuracy.

相似文献/References:

[1]段逸凡,肖洪兵,Rahman Md Mostafizur.基于3DSEU-Net不确定性循环焦点平均教师的半监督脑肿瘤分割[J].中国医学物理学杂志,2023,40(9):1121.[doi:DOI:10.3969/j.issn.1005-202X.2023.09.010]
　DUAN Yifan,XIAO Hongbing,Rahman Md Mostafizur.Semi-supervised learning for brain tumor segmentation through 3DSEU-Net as uncertainty-aware mean teacher and cyclical focal loss[J].Chinese Journal of Medical Physics,2023,40(1):1121.[doi:DOI:10.3969/j.issn.1005-202X.2023.09.010]
[2]邹祥,王瑜,肖洪兵,等.基于有效感受野和注意力融合机制的脑肿瘤全自动分割[J].中国医学物理学杂志,2024,41(5):563.[doi:DOI:10.3969/j.issn.1005-202X.2024.05.006]
　ZOU Xiang,WANG Yu,XIAO Hongbing,et al.Fully-automatic brain tumor segmentation based on effective receptive field and attention fusion mechanism[J].Chinese Journal of Medical Physics,2024,41(1):563.[doi:DOI:10.3969/j.issn.1005-202X.2024.05.006]
[3]刘恒,朱俊杰.PAC-UNet：融合双路径卷积与注意力机制的脑肿瘤分割网络[J].中国医学物理学杂志,2026,43(6):732.[doi:DOI:10.3969/j.issn.1005-202X.2026.06.005]
　brain tumor segmentation dual-path convolution pinwheel convolution dynamic kernel convolution attention mechanism U-Net.PAC-UNet: a brain tumor segmentation network integrating dual-path convolution and attention mechanism[J].Chinese Journal of Medical Physics,2026,43(1):732.[doi:DOI:10.3969/j.issn.1005-202X.2026.06.005]

备注/Memo

备注/Memo:
【收稿日期】2025-07-19 【基金项目】北京市教育委员会-北京市自然基金联合项目（KZ202110 011015） 【作者简介】李国鹏，硕士研究生，研究方向：图像处理与模式识别，E-mail: stulgp@163.com 【通信作者】肖洪兵，副教授，研究方向：图像处理与模式识别，E-mail: x.hb@163.com
更新日期/Last Update: 2026-01-27