北京大学第一医院太原医院/太原市中心医院(太原 030009)
王军。E-mail:wangjun8998@163.com
收稿:2025-10-05,
修回:2026-02-03,
网络首发:2026-08-17,
纸质出版:2026-08-20
移动端阅览
周丰艺,杨喆,马苑,等.CB-SwinUNet:一种融合多模块注意力机制的乳腺超声肿瘤分割算法[J].北京生物医学工程,2026,45(4):376-385.
ZHOU Fengyi,YANG Zhe,MA Yuan,et al.CB-SwinUNet: a multi-module attention fusion algorithm for breast ultrasound tumor segmentation[J].Beijing Biomedical Engineering,2026,45(4):376-385.
周丰艺,杨喆,马苑,等.CB-SwinUNet:一种融合多模块注意力机制的乳腺超声肿瘤分割算法[J].北京生物医学工程,2026,45(4):376-385. DOI: 10.3969/j.issn.1002-3208.2026.04.007.
ZHOU Fengyi,YANG Zhe,MA Yuan,et al.CB-SwinUNet: a multi-module attention fusion algorithm for breast ultrasound tumor segmentation[J].Beijing Biomedical Engineering,2026,45(4):376-385. DOI: 10.3969/j.issn.1002-3208.2026.04.007.
目的
2
针对乳腺超声图像中强散斑噪声、边界模糊及低对比度等问题,现有方法在处理不规则肿瘤与复杂背景时,难以兼顾全局语义与局部细节。为此,本文提出融合Swin Transformer与U-Net优势的CB-SwinUNet模型,旨在实现复杂场景下的高精度分割。
方法
2
模型构建基于Swin Transformer的U型架构,利用移动窗口自注意力捕捉长距离依赖。在此基础上引入两个增强模块:在编码阶段并行引入通道注意力与空间增强模块(channel attention and spatial enhancement, CASE),联合建模空间与通道特征,自适应抑制背景噪声并增强病灶表达;在跳跃连接处设计边界细节增强模块(boundary-aware detail module, BADM),利用轻量级卷积与多尺度池化聚合浅层高频信息,修正深层特征的边缘丢失。此外,结合复合损失函数与患者级五折交叉验证策略确保评估客观性。
结果
2
在BUSI(647例)与自建数据库(500例)上的实验显示,CB-SwinUNet的Dice系数分别达到0.859和0.842,IoU与Hausdorff距离等指标均优于U-Net、TransFuse等主流模型。消融实验证实,CASE显著提升了低对比度病灶定位能力,BADM有效修正了边缘刻画精度。模型收敛迅速且未出现明显过拟合。
结论
2
CB-SwinUNet通过深度嵌入式融合策略,实现了全局结构与局部细节的协同建模,有效解决了噪声干扰与边界模糊难题。实验结果充分验证了该方法在提升分割精度与泛化能力方面的优势,为临床辅助诊断提供了鲁棒的自动化方案。
Objective
2
Existing methods often struggle to balance global semantic understanding with local detail recovery when dealing with irregular tumor shapes and complex backgrounds, due to prevalent issues in breast ultrasound images such as severe speckle noise, blurred boundaries, and low contrast. To address these challenges, this paper proposes a CB-SwinUNet model that integrates the advantages of Swin Transformer and U-Net, aiming to achieve high-precision segmentation in complex clinical scenarios.
Methods
2
The model constructs a U-shaped architecture based on the Swin Transformer, utilizing shifted window self-attention mechanisms to capture long-range dependencies. Building on this, two key enhancement modules are introduced. First, a channel attention and spatial enhancement (CASE) module is incorporated in parallel during the encoding stage to jointly model spatial and channel features, adaptively suppressing background noise and enhancing lesion representation. Second, a boundary-aware detail module (BADM) is designed at the skip connections, leveraging lightweight convolutions and multi-scale pooling to aggregate shallow high-frequency information, thereby correcting the loss of edge details in deep features. Furthermore, a composite loss function and a patient-level 5-fold cross-validation strategy are employed to ensure objective evaluation.
Results
2
Experiments on the BUSI dataset (647 cases) and a self-built database (500 cases) demonstrate that CB-SwinUNet achieves Dice coefficients of 0.859 and 0.842, respectively. Metrics such as intersection over union (IoU) and Hausdorff distance are superior to those of mainstream models like U-Net and TransFuse. Ablation studies confirm that CASE significantly improves the localization capability for low-contrast lesions, while BADM effectively refines edge delineation accuracy. The model exhibits rapid convergence without significant overfitting.
Conclusions
2
Through a deep embedded fusion strategy, CB-SwinUNet achieves synergistic modeling of global structures and local details, effectively resolving challenges related to noise interference and blurred boundaries. The experimental results fully validate the advantages of the proposed method in enhancing segmentation accuracy and generalization ability, providing a robust automated solution for clinical computer-aided diagnosis.
Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA: A Cancer Journal for Clinicians, 2021, 71(3): 209-249.
Iacob R, Iacob ER, Stoicescu ER, et al. Evaluating the role of breast ultrasound in early detection of breast cancer in low-and middle-income countries: a comprehensive narrative review[J]. Bioengineering, 2024, 11(3): 262.
Abbott JG, Thurstone FL. Acoustic speckle: theory and experimental analysis[J]. Ultrasonic imaging, 1979, 1(4): 303-324.
Doi K, Giger ML, Nishikawa RM, et al. Computer-aided diagnosis of breast cancer on mammograms[J]. Breast Cancer, 1997, 4(4): 228-233.
Siddique N, Paheding S, Elkin CP, et al. U-net and its variants for medical image segmentation: a review of theory and applications[J]. IEEE Access, 2021, 9: 82031-82057.
Liu LL, Wang Y, Chang J, et al. LLRHNet: multiple lesions segmentation using local-long range features[J]. Frontiers in Neuroinformatics, 2022, 16: 859973.
Feng JL, Dong XX, Liu XJ, et al. HTBE-Net: a hybrid transformer network based on boundary enhancement for breast ultrasound image segmentation[J]. Displays, 2024, 84: 102753.
Shamshad F, Khan S, Zamir SW, et al. Transformers in medical imaging: a survey[J]. Medical Image Analysis, 2023, 88: 102802.
Otsu N. A threshold selection method from gray-level histograms[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1): 62-66.
Adams R, Bischof L. Seeded region growing[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1994, 16(6): 641-647.
Kass M, Witkin A, Terzopoulos D. Snakes: active contour models[J]. International Journal of Computer Vision, 1988, 1: 321-331.
Zhou ZW, Siddiquee MMR, Tajbakhsh N, et al. Unet++: a nested u-net architecture for medical image segmentation[C]// Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support(2018). Granada, Spain: DLMIA, 2018, 11045: 3-11.
Woo S, Park J, Lee JY, et al. Cbam: convolutional block attention module[C]// European Conference on Computer Vision (ECCV) 2018. Munich, Germany: Springer, Cham, 2018: 3-19.
Jha D, Smedsrud PH, Riegler MA, et al. ResUNet++: an advanced architecture for medical image segmentation[C]// 2019 IEEE International Symposium on Multimedia (ISM). San Diego, CA, USA: IEEE Press: 2019: 225-2255.
Punn NS, Agarwal S. RCA-IUnet: a residual cross-spatial attention-guided inception U-Net model for tumor segmentation in breast ultrasound imaging[J]. Machine Vision and Applications, 2022, 33: 27.
Lou M, Meng J, Qi YL, et al. MCRNet: multi-level context refinement network for semantic segmentation in breast ultrasound imaging[J]. Neurocomputing, 2022, 470: 154-169.
Liu Z, Lin YT, Cao Y, et al. Swin Transformer: hierarchical vision transformer using shifted windows[C]// 2021 IEEE/CVF International Conference on Computer Vision(ICCV). Montreal, QC, Canada: IEEE Press, 2021: 9992-10002.
Hatamizadeh A, Tang YC, Nath V, et al. UNETR: Transformers for 3D medical image segmentation[C]// 2022 IEEE/CVF Winter Conference on Applications of Computer Vision(WACV). Waikoloa, HI, USA: IEEE Press, 2022: 1748-1758.
Wang C, Li H, Zhang Y, et al. Sclera-TransFuse: Fusing vision transformer and CNN for accurate sclera segmentation and recognition[J]. IEEE Transactions on Biometrics, Behavior, and Identity Science, 2024, 6(4): 575-590.
Pan Y, Chen Q, Fang X. DAMAF: dual attention network with multi-level adaptive complementary fusion for medical image segmentation[J]. The Visual Computer, 2025, 41: 2409-2424.
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621