[1]李柯,刘文忠,秦镜淘.改进MambaUNet网络对肝脏肿瘤CT图像的轻量化级联分割[J].中国医学物理学杂志,2025,42(8):1068-1078.[doi:DOI:10.3969/j.issn.1005-202X.2025.08.014]
　LI Ke,LIU Wenzhong,QIN Jingtao,et al.Improved MambaUNet for lightweight cascaded segmentation in liver tumor CT image[J].Chinese Journal of Medical Physics,2025,42(8):1068-1078.[doi:DOI:10.3969/j.issn.1005-202X.2025.08.014]
点击复制

改进MambaUNet网络对肝脏肿瘤CT图像的轻量化级联分割()

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

卷:
42
期数:
2025年第8期
页码:
1068-1078
栏目:
医学影像物理
出版日期:
2025-08-30

文章信息/Info

Title:
Improved MambaUNet for lightweight cascaded segmentation in liver tumor CT image
文章编号:
1005-202X（2025）08-1068-11
作者:
李柯1; 刘文忠1; 2; 秦镜淘1
1.四川轻化工大学计算机科学与工程学院， 四川 宜宾 644002； 2.四川省企业信息化与物联网高等教育重点实验室， 四川 宜宾 644002
Author(s):
LI Ke1;  LIU Wenzhong1;  2;  QIN Jingtao1
1. School of Computer Science and Engineering, Sichuan University of Science and Engineering, Yibin 644002, China 2. Key Laboratory of Higher Education of Sichuan Province for Enterprise Informationalization and Internet of Things, Yibin 644002, China
关键词:
肝脏肿瘤分割; MD-MambaUNet; CT图像; Dice相似性系数
Keywords:
Keywords: liver tumor segmentation MD-MambaUNet CT image Dice similarity coefficient
分类号:
R318；TP391.4
DOI:
DOI:10.3969/j.issn.1005-202X.2025.08.014
文献标志码:
A
摘要:
为解决卷积神经网络在全局上下文建模能力的局限性以及Transformer在自注意力机制中二次计算的复杂性，提出一种基于MambaUNet改进的MD-MambaUNet网络。该网络结合了多向选择扫描模块（MD-SS2D），从CT图像的多个方向提取空间特征，显著提升全局上下文建模能力，并通过引入部分卷积，构建轻量化混合卷积模块，显著减少模型的参数量，能在保持高水平性能的同时，以更低的计算成本处理大规模的医疗图像数据。基于LiTS2017、3DIRCADB公共数据集进行实验，MD-MambaUNet在LiTS2017数据集中，肝脏与肿瘤分割的Dice相似性系数指标分别较MambaUNet网络提高3.32%和4.77%，达到95.36%和76.93；交并比指标分别较MambaUNet网络提高4.18%和4.92%，达到91.43%和69.74%。在3DIRCADB数据集中，肝脏与肿瘤分割的Dice相似性系数指标分别较MambaUNet网络提高2.08%和2.21%，达到93.81%和64.68%；交并比指标分别较MambaUNet网络提高0.79%和3.13%，达到87.23%和57.65%，同时参数量相比MambaUNet减少了13.71 M，使得模型能部署在临床场景成为可能。
Abstract:
Abstract: To address the limitations of convolutional neural networks in global context modeling and the complexity of secondary computation in Transformers self attention mechanism, an improved multi-direction-MambaUNet (MD-MambaUNet) based on MambaUNet is proposed. This network integrates with multi directional selective scanning module to extract spatial features from multiple directions of CT images, significantly improving the global context modeling capability. By introducing partial convolution and constructing a lightweight hybrid convolution module, the models parameter count is significantly reduced, enabling the processing of large-scale medical image data at lower computational costs while maintaining high-level performance. Experiments are conducted on the LiTS2017 and 3DIRCADB public datasets. Compared with MambaUNet on the LiTS2017 dataset, MD-MambaUNet improves the Dice similarity coefficient for liver and tumor segmentation by 3.32% and 4.77%, reaching 95.36% and 76.93%, and increases intersection over union by 4.18% and 4.92%, reaching 91.43% and 69.74%, respectively. Compared with MambaUNet on the 3DIRCADB dataset, MD-MambaUNet improves the Dice similarity coefficient for liver and tumor segmentation by 2.08% and 2.21%, reaching 93.81% and 64.68%, and increases intersection over union by 0.79% and 3.13%, reaching 87.23% and 57.65%, respectively. Meanwhile, the parameter count is 13.71 M less than MambaUNet, making it possible for the model to be deployed in clinical scenarios.

相似文献/References:

[1]潘依乐,高永彬.多模态弱监督学习在肝癌图像生成与分割中的应用[J].中国医学物理学杂志,2024,41(1):8.[doi:DOI:10.3969/j.issn.1005-202X.2024.01.002]
　PAN Yile,GAO Yongbin.Application of multimodal weakly-supervised learning in image synthesis and segmentation of liver cancer[J].Chinese Journal of Medical Physics,2024,41(8):8.[doi:DOI:10.3969/j.issn.1005-202X.2024.01.002]
[2]扈蕴琨,王晓艳,王秀娟.基于级联DDR-UNet++的肝脏肿瘤图像分割方法[J].中国医学物理学杂志,2025,42(7):901.[doi:DOI:10.3969/j.issn.1005-202X.2025.07.009]
　HU Yunkun,WANG Xiaoyan,WANG Xiujuan.Liver tumor image segmentation method based on cascaded DDR-UNet++[J].Chinese Journal of Medical Physics,2025,42(8):901.[doi:DOI:10.3969/j.issn.1005-202X.2025.07.009]

备注/Memo

备注/Memo:
【收稿日期】2025-02-12 【基金项目】四川轻化工大学人才引进项目（2018RCL20） 【作者简介】李柯，硕士研究生，研究方向：医学图像处理、数字图像处理，E-mail: like6715@foxmail.com 【通信作者】刘文忠，副教授，研究方向：生物信息学、生物医学图像处理，E-mail: liuwz@suse.edu.cn
更新日期/Last Update: 2025-09-15