基于2.5D Attention U-Net的颅脑CT图像分割研究
Research on Craniocerebral CT Image Segmentation Based on 2.5D Attention U-Net
DOI: 10.12677/csa.2026.169285, PDF,    科研立项经费支持
作者: 郭靖宇, 陈文轩：应急管理大学，计算机与信息安全学院，北京；张云雷*：应急管理大学，计算机与信息安全学院，北京；河北省物联网监控技术创新中心，北京
关键词: 颅脑CT；图像分割；2.5D；Attention U-Net；ResNet34；深度学习；Brain CT； Image Segmentation； 2.5D； Attention U-Net； ResNet34； Deep Learning
摘要: 颅脑CT图像中目标区域的自动分割能够为病灶定位、体积测量和后续定量分析提供基础。由于CT图像存在灰度差异小、边界不清、目标面积占比低以及空标签切片较多等问题，常规二维分割网络容易出现漏检或边界缺失。针对上述问题，本文构建一种2.5D Attention U-Net分割模型。该模型以当前切片及其前后相邻切片作为三通道输入，在保留二维卷积计算效率的同时引入层间空间信息；在U-Net跳跃连接处加入通道与空间联合注意力模块，对编码端特征进行自适应加权；并采用GroupNorm提高小批量训练时的稳定性。训练阶段使用二元交叉熵与Tversky损失的组合，同时在验证集上选择预测阈值，以减轻前景与背景不平衡对模型选择的影响。为检验模型效果，本文设置单切片ResNet34-U-Net作为对照，并采用Dice、交并比、精确率、召回率及空切片误报率进行评价。本研究可为小目标、弱边界颅脑CT图像的自动分割提供一种计算量与层间信息利用相兼顾的实现方案。
Abstract: Automatic segmentation of target regions in brain CT images provides a foundation for lesion localization, volume measurement, and subsequent quantitative analysis. However, brain CT images often exhibit small grayscale differences, unclear boundaries, a low target-area proportion, and a large number of slices with empty labels. These issues can lead conventional two-dimensional segmentation networks to miss targets or produce incomplete boundaries. To address these problems, this study developed a 2.5D Attention U-Net segmentation model. The model uses the current slice together with its preceding and following slices as a three-channel input, thereby introducing inter-slice spatial information while retaining the computational efficiency of two-dimensional convolution. A combined channel and spatial attention module is incorporated into the skip connections of U-Net to adaptively reweight encoder features. Group Normalization is employed to improve training stability under small-batch conditions. During training, a combination of binary cross-entropy loss and Tversky loss is used. The prediction threshold is selected on the validation set to reduce the influence of foreground-background imbalance on model selection. To evaluate the model, a single-slice ResNet34-U-Net is used as a comparison model. Dice, Intersection over Union, precision, recall, and the false-positive rate on empty slices are used as evaluation metrics. This study provides an implementation approach for automatic segmentation of small-target and weak-boundary regions in brain CT images, balancing the use of inter-slice information with computational efficiency.
文章引用：郭靖宇, 张云雷, 陈文轩. 基于2.5D Attention U-Net的颅脑CT图像分割研究[J]. 计算机科学与应用, 2026, 16(9): 14-24. https://doi.org/10.12677/csa.2026.169285

参考文献

[1] Ronneberger, O., Fischer, P. and Brox, T. (2015) U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab, N., Hornegger, J., Wells, W. and Frangi, A., Eds., Medical Image Computing and Computer-Assisted Intervention—MICCAI 2015, Springer, 234-241.
https://doi.org/10.1007/978-3-319-24574-4_28
[2] Oktay, O., Schlemper, J., Le Folgoc, L., et al. (2018) Attention U-Net: Learning Where to Look for the Pancreas. arXiv: 1804.03999.
[3] Zhou, Z., Siddiquee, M.M.R., Tajbakhsh, N. and Liang, J. (2020) UNet++: Redesigning Skip Connections to Exploit Multiscale Features in Image Segmentation. IEEE Transactions on Medical Imaging, 39, 1856-1867.
https://doi.org/10.1109/tmi.2019.2959609
[4] Isensee, F., Jaeger, P.F., Kohl, S.A.A., Petersen, J. and Maier-Hein, K.H. (2020) nnU-Net: A Self-Configuring Method for Deep Learning-Based Biomedical Image Segmentation. Nature Methods, 18, 203-211.
https://doi.org/10.1038/s41592-020-01008-z
[5] Ma, J., Chen, J., Ng, M., Huang, R., Li, Y., Li, C., et al. (2021) Loss Odyssey in Medical Image Segmentation. Medical Image Analysis, 71, Article ID: 102035.
https://doi.org/10.1016/j.media.2021.102035
[6] He, K., Zhang, X., Ren, S. and Sun, J. (2016) Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, 27-30 June 2016, 770-778.
https://doi.org/10.1109/cvpr.2016.90
[7] Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B., et al. (2022) UNETR: Transformers for 3D Medical Image Segmentation. 2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Waikoloa, 3-8 January 2022, 1748-1758.
https://doi.org/10.1109/wacv51458.2022.00181
[8] Ma, J., He, Y., Li, F., Han, L., You, C. and Wang, B. (2024) Segment Anything in Medical Images. Nature Communications, 15, Article No. 654.
https://doi.org/10.1038/s41467-024-44824-z
[9] Xing, Z., Ye, T., Yang, Y., Liu, G. and Zhu, L. (2024) SegMamba: Long-Range Sequential Modeling Mamba for 3D Medical Image Segmentation. In: Linguraru, M.G., et al., Eds., Medical Image Computing and Computer Assisted Intervention—MICCAI 2024, Springer, 578-588.
https://doi.org/10.1007/978-3-031-72111-3_54
[10] MacIntosh, B.J., Liu, Q., Schellhorn, T., Beyer, M.K., Groote, I.R., Morberg, P.C., et al. (2023) Radiological Features of Brain Hemorrhage through Automated Segmentation from Computed Tomography in Stroke and Traumatic Brain Injury. Frontiers in Neurology, 14, Article 1244672.
https://doi.org/10.3389/fneur.2023.1244672
[11] Lin, E. and Yuh, E.L. (2024) Semi-Supervised Learning for Generalizable Intracranial Hemorrhage Detection and Segmentation. Radiology: Artificial Intelligence, 6, e230077.
https://doi.org/10.1148/ryai.230077
[12] Hoang, Q.T., Pham, X.H., Trinh, X.T., Le, A.V., Bui, M.V. and Bui, T.T. (2024) An Efficient CNN-Based Method for Intracranial Hemorrhage Segmentation from Computerized Tomography Imaging. Journal of Imaging, 10, Article 77.
https://doi.org/10.3390/jimaging10040077
[13] Ma, D., Li, C., Du, T., Qiao, L., Tang, D., Ma, Z., et al. (2024) PHE-SICH-CT-IDS: A Benchmark CT Image Dataset for Evaluation Semantic Segmentation, Object Detection and Radiomic Feature Extraction of Perihematomal Edema in Spontaneous Intracerebral Hemorrhage. Computers in Biology and Medicine, 173, Article ID: 108342.
https://doi.org/10.1016/j.compbiomed.2024.108342
[14] Hu, P., Yan, T., Xiao, B., Shu, H., Sheng, Y., Wu, Y., et al. (2024) Deep Learning-Assisted Detection and Segmentation of Intracranial Hemorrhage in NonContrast Computed Tomography Scans of Acute Stroke Patients: A Systematic Review and Meta-Analysis. International Journal of Surgery, 110, 3839-3847.
https://doi.org/10.1097/js9.0000000000001266
[15] Li, Y., Zhang, R., Li, Y., Zuo, X., Wang, Q., Zhang, S., et al. (2024) Deep Learning Models for Separate Segmentations of Intracerebral and Intraventricular Hemorrhage on Head CT and Segmentation Quality Assessment. Medical Physics, 51, 8317-8333.
https://doi.org/10.1002/mp.17343