[1]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):732-740.[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(6):732-740.[doi:DOI:10.3969/j.issn.1005-202X.2026.06.005]
点击复制

PAC-UNet：融合双路径卷积与注意力机制的脑肿瘤分割网络()

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

卷:
43卷
期数:
2026年第6期
页码:
732-740
栏目:
医学影像物理
出版日期:
2026-06-26

文章信息/Info

Title:
PAC-UNet: a brain tumor segmentation network integrating dual-path convolution and attention mechanism
文章编号:
1005-202X（2026）06-0732-09
作者:
刘恒1; 朱俊杰1; 2
1.河南理工大学电气工程与自动化学院， 河南 焦作 454003； 2.河南省煤矿装备智能检测与控制重点实验室， 河南 焦作 454003
Author(s):
brain tumor segmentation dual-path convolution pinwheel convolution dynamic kernel convolution attention mechanism U-Net
1. School of Electrical Engineering and Automation, Henan Polytechnic University, Jiaozuo 454003, China 2. Henan Key Laboratory of Intelligent Detection and Control of Coal Mine Equipment, Jiaozuo 454003, China
关键词:
脑肿瘤分割; 双路径卷积; 风车状卷积; 动态核卷积; 注意力机制; U-Net
Keywords:
brain tumor segmentation dual-path convolution pinwheel convolution dynamic kernel convolution attention mechanism U-Net
分类号:
R318
DOI:
DOI:10.3969/j.issn.1005-202X.2026.06.005
文献标志码:
A
摘要:
针对脑肿瘤MRI影像中病灶边界模糊、小目标分割敏感性不足及多尺度上下文信息捕获不充分等问题，提出一种融合双路径卷积模块与混合注意力机制模块的U型分割网络PAC-UNet。该网络采用新型双路径卷积模块替换标准卷积，通过并行结构整合动态核卷积与风车状卷积路径，其中，动态核卷积路径结合径向基函数RBF核生成器替代传统固定卷积核，通过可微分样条参数化实现核函数的动态适应，平衡局部细节特征与全局语义表征；风车状卷积路径采用四并行结构有效扩展感受野，增强上下文信息捕获能力。同时利用通道与空间双重注意力机制融合双路径特征，聚焦关键特征并抑制冗余背景干扰，协同优化多尺度特征表达。在BraTS2023数据集上的实验结果表明，PAC-UNet能够显著提升脑肿瘤分割性能。
Abstract:
To address the issues of blurred lesion boundaries in brain tumor MRI images, insufficient segmentation sensitivity for small targets, and limited ability to extract multi-scale contextual information, this study proposes a U-shaped segmentation network named PAC-UNet, which integrates a dual-path convolution module and a hybrid attention mechanism module. The network uses a novel dual-path convolution module to replace standard convolutions, enabling parallel fusion of dynamic kernel convolution and pinwheel convolution paths. In the dynamic kernel convolution path, a radial basis function kernel generator replaces traditional fixed convolution kernels, which achieves dynamic kernel function adaptation through differentiable spline parameterization, thus balancing local detailed features and global semantic representations. The pinwheel convolution path adopts a four-parallel structure to effectively expand the receptive field and enhance the ability to capture contextual information. Meanwhile, a channel-spatial dual attention mechanism is used to fuse dual-path features, with a focus on key features while suppressing redundant background interference, thereby collaboratively optimizing multi-scale feature representations. Experiment results on the BraTS2023 dataset show that PAC-UNet achieves significant improvements in brain tumor segmentation performance.

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(6):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(6):563.[doi:DOI:10.3969/j.issn.1005-202X.2024.05.006]
[3]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.[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(6):76.[doi:DOI:10.3969/j.issn.1005-202X.2026.01.010]

备注/Memo

备注/Memo:
【收稿日期】2026-02-20 【基金项目】国家自然科学基金（61601173）；河南理工大学博士创新基金（760807/013） 【作者简介】刘恒，硕士研究生，研究方向：基于深度学习的医学图像处理，E-mail: lh8615@home.hpu.edu.cn 【通信作者】朱俊杰，博士，研究方向：深度学习、信号分析、生物医学信号处理、源重建及磁心电成像，E-mail: junjiezhu@hpu.edu.cn
更新日期/Last Update: 2026-06-26