SSCU-UNet：融合空间位移与内容感知注意力的医学图像分割模型

SSCU-UNet: a medical image segmentation network integrating spatial shift and content-aware attention

摘要

目的：设计一种能兼具全局感知与高分辨率特征保持能力的分割框架SSCU-UNet，以提升多器官CT与心脏MRI的分割性能。方法：提出SSCU-UNet分割框架，包含3个核心设计：（1）残差卷积下采样模块采用双路径结构，通过Conv-GELU-Conv主路径与步长卷积快捷路径融合，并结合层归一化以保留高分辨率空间特征。（2）SSFormer模块，位于每个Swin Transformer块中的窗口注意力和多层感知器之间，沿残差分支运行，以增强局部空间建模，同时通过空间位移注意力保持全局上下文推理。（3）内容感知注意力上采样模块根据输入特征自适应生成重建权重，以实现更精确的边界恢复。结果：在Synapse数据集上，SSCU-UNet的平均Dice相似系数为82.62%，较U-Net、TransUNet与Swin-UNet分别提升5.77%、5.14%和3.49%，Hausdorff距离降至18.79 mm，显著改善胆囊、胰腺、胃等形态复杂器官的边界精度。在ACDC数据集上，平均Dice相似系数达92.34%，优于多种基线模型。结论：SSCU-UNet通过整合残差卷积下采样、空间位移增强局部建模和内容感知上采样，有效平衡全局语义推理与细粒度结构恢复，在多器官分割任务中取得先进性能。

Abstract

Objective To design a segmentation framework, namely spatial shift-enhanced and content-aware upsampling UNet (SSCU-UNet), which balances global context awareness and high-resolution feature retention to improve performance in multi-organ CT and cardiac MRI segmentation tasks. Methods The proposed SSCU-UNet segmentation framework incorporated 3 key components. (1) A residual convolutional downsampling module adopted a dual-branch structure consisting of a Conv-GELU-Conv main branch and a strided-convolution shortcut branch, and coupled with layer normalization to retain high-resolution spatial information. (2) A SSFormer module was inserted between the window attention and multilayer perceptron of each Swin Transformer block, and operates along the residual branch to enhance local spatial modeling while maintaining global contextual reasoning via spatial shift attention. (3) A content-aware attention upsampling module was used to enable finer boundary restoration, which adaptively generates feature reconstruction weights according to input features. Results The proposed SSCU-UNet achieved an average Dice similarity coefficient of 82.62% on the Synapse dataset, which was 5.77%, 5.14%, and 3.49% higher than U-Net, TransUNet, and Swin-UNet, respectively, and reduced the Hausdorff distance to 18.79 mm, with particularly prominent performance improvements for anatomically complex organs including the gallbladder, pancreas, and stomach. On the ACDC dataset, the SSCU-UNet yielded an average Dice similarity coefficient of 92.34%, outperforming multiple baseline models. Conclusion By integrating residual convolutional downsampling, spatial shift-enhanced local modeling, and content-aware upsampling, the SSCU-UNet effectively balances global semantic reasoning with fine-grained structural reconstruction, and achieves state-of-the-art performance in multi-organ segmentation tasks.

关键词

医学图像分割 / SSCU-UNet / Swin Transformer / 空间位移模块 / 内容感知上采样 / 多器官分割

Key words

medical image segmentation / spatial shift-enhanced and content-aware upsampling UNet / Swin Transformer / spatial shift module / content-aware upsampling / multi-organ segmentation

参考文献

[1] Wang JJ, Wang SH, Zhang YD . Deep learning on medical image analysis[J]. CAAI Trans Intell Technol, 2025, 10(1): 1-35.

[2] Chen XX, Wang XM, Zhang K, et al. Recent advances and clinical applications of deep learning in medical image analysis[J]. Med Image Anal, 2022, 79: 102444.

[3] Ronneberger O, Fischer P, Brox T . U—Net: convolutional networks for biomedical image segmentation[C]// Medical Image Computing and Computer—Assisted Intervention—MICCAI 2015. Cham: Springer International Publishing, 2015: 234-241.

[4] Zhou ZW, Rahman Siddiquee MM, 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. Cham: Springer International Publishing, 2018: 3-11.

[5] Diakogiannis FI, Waldner F, Caccetta P, et al. ResUNet—a: a deep learning framework for semantic segmentation of remotely sensed data[J]. ISPRS J Photogramm Remote Sens, 2020, 162: 94-114.

[6] Huang HM, Lin LF, Tong RF, et al. UNet 3+: a full—scale connected UNet for medical image segmentation[C]// ICASSP 2020 — 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Piscataway, NJ, USA: IEEE, 2020: 1055-1059.

[7] Ibtehaz N, Rahman MS . MultiResUNet: rethinking the U—Net architecture for multimodal biomedical image segmentation[J]. Neural Netw, 2020, 121: 74-87.

[8] Jha D, Riegler MA, Johansen D, et al. DoubleU—Net: a deep convolutional neural network for medical image segmentation[C]// 2020 IEEE 33rd International Symposium on Computer—Based Medical Systems (CBMS). Piscataway, NJ, USA: IEEE, 2020: 558-564.

[9] Oktay O, Schlemper J, Le Folgoc L, et al. Attention U—Net: learning where to look for the pancreas[EB/OL]. (2018—05—20)[2026—04—29]. https://arxiv.org/abs/1804.03999.

[10] Isensee F, Jaeger PF, Kohl SA, et al. nnU—Net: a self—configuring method for deep learning—based biomedical image segmentation[J]. Nat Methods, 2021, 18(2): 203-211.

[11] Çiçek Ö, Abdulkadir A, Lienkamp SS, et al. 3D U—Net: learning dense volumetric segmentation from sparse annotation[C]// Medical Image Computing and Computer—Assisted Intervention—MICCAI 2016. Cham: Springer International Publishing, 2016: 424-432.

[12] Milletari F, Navab N, Ahmadi SA . V—Net: fully convolutional neural networks for volumetric medical image segmentation[C]// 2016 Fourth International Conference on 3D Vision (3DV). Piscataway, NJ, USA: IEEE, 2016: 565-571.

[13] Dai DW, Dong CX, Yan QS, et al. I2U—Net: A dual—path U—Net with rich information interaction for medical image segmentation[J]. Med Image Anal, 2024, 97: 103241.

[14] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., 2017: 6000-6010.

[15] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. (2021—06—03)[2026—04—29]. https://arxiv.org/abs/2010.11929.

[16] Chen JN, Lu YY, Yu QH, et al. TransUNet: transformers make strong encoders for medical image segmentation[EB/OL]. (2021—02—8)[2026—04—29]. https://arxiv.org/abs/2102.04306.

[17] Cao H, Wang YY, Chen J, et al. Swin—Unet: Unet—like pure transformer for medical image segmentation[C]// Computer Vision—ECCV 2022 Workshops. Cham: Springer Nature Switzerland, 2023: 205-218.

[18] Liu X, Gao P, Yu T, et al. CSWin—UNet: transformer UNet with cross—shaped windows for medical image segmentation[J]. Inf Fusion, 2025, 113: 102634.

[19] Valanarasu JM, Oza P, Hacihaliloglu I, et al. Medical transformer: gated axial—attention for medical image segmentation[C]// Medical Image Computing and Computer Assisted Intervention—MICCAI 2021. Cham: Springer International Publishing, 2021: 36-46.

[20] Hatamizadeh A, Tang YC, Nath V, et al. UNETR: transformers for 3D medical image segmentation[EB/OL]. (2021—10—09)[2026—04—29]. https://arxiv.org/abs/2103.10504.

[21] Xie YT, Zhang JP, Shen CH, et al. CoTr: efficiently bridging CNN and transformer for 3D medical image segmentation[C]// Medical Image Computing and Computer Assisted Intervention—MICCAI 2021. Cham: Springer International Publishing, 2021: 171-180.

[22] Jiang Y, Zhang Y, Lin X, et al. SwinBTS: a method for 3D multimodal brain tumor segmentation using Swin transformer[J]. Brain Sci, 2022, 12(6): 797.

[23] Huang XH, Deng ZF, Li DD, et al. MISSFormer: an effective transformer for 2D medical image segmentation[J]. IEEE Trans Med Imaging, 2023, 42(5): 1484-1494.

[24] Heidari M, Kazerouni A, Soltany M, et al. HiFormer: hierarchical multi—scale representations using transformers for medical image segmentation[EB/OL]. (2023—01—09)[2026—04—29]. https://arxiv.org/abs/2207.08518.

[25] Ren SC, Zhou DQ, He SF, et al. Shunted self—attention via multi—scale token aggregation [C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway, NJ, USA: IEEE, 2022: 10843-10852.

[26] Fu LY, Chen YZ, Ji W, et al. SSTrans—Net: smart Swin transformer network for medical image segmentation[J]. Biomed Signal Process Control, 2024, 91: 106071.

[27] Çetiner H, Metlek S . DenseUNet+: a novel hybrid segmentation approach based on multi—modality images for brain tumor segmentation[J]. J King Saud Univ Comput Inf Sci, 2023, 35(8): 101663.

[28] Xing ZH, Wan L, Fu HZ, et al. Diff—UNet: a diffusion embedded network for volumetric segmentation[EB/OL]. (2023—03—18)[2026—04—29]. https://arxiv.org/abs/2303.10326.

[29] Wang JH, Chen JT, Chen D, et al. LKM—UNet: large kernel vision mamba UNet for medical image segmentation[C]// Medical Image Computing and Computer Assisted Intervention—MICCAI 2024. Cham: Springer Nature Switzerland, 2024: 360-370.

[30] Wang JQ, Chen K, Xu R, et al. CARAFE: content—aware reassembly of features[C]// 2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway, NJ, USA: IEEE, 2019: 3007-3016.

[31] Fu SH, Lu YY, Wang Y, et al. Domain adaptive relational reasoning for 3D multi—organ segmentation[C]// Medical Image Computing and Computer Assisted Intervention—MICCAI 2020. Cham: Springer International Publishing, 2020: 656-666.

[32] Ruan JC, Li JC, Xiang SC . VM—UNet: vision mamba UNet for medical image segmentation[J/OL]. ACM Transactions on Multimedia Computing, Communications and Applications. (2025—09—16)[2026—04—29]. https://doi.org/10.1145/3767748.

基金资助

泉州市科技项目(2023N002S)