先进急诊医学

深度学习框架下的医学图像分割算法研究

张 婷玮

辽宁科技大学

龙 艳彬

辽宁科技大学

邓 凯文

辽宁科技大学

DOI: https://doi.org/10.59429/xjjz.v6i4.8878

Keywords: 医学图像分割；U-Net；Transformer；Mamba；分割一切模型（SAM）

---

Abstract

针对医学图像的高维性、复杂性和高精度要求，论文综述了深度学习框架下的医学图像分割算法。深度学 习模型能够自适应地从大量数据中学习并提取多层次特征，为医学图像分割提供了高精度、高鲁棒性和可扩展性强 的解决方案。论文重点讨论了 U-Net、Transformer、Mamba 以及分割一切模型（SAM）等先进模型在医学图像分割 中的应用，并从多个维度进行了综合对比分析。此外，论文还总结了当前医学图像分割研究面临的挑战，并对未来 研究方向进行了展望。

---

References

[1] HSU C Y, TSENG C C, LEE S L, et al. Image classificationusing convolutional neural networks with differentconvolution operations[C/OL]//2020 IEEE InternationalConference on Consumer Electronics-Taiwan (ICCE-Taiwan),2020:1-2.

[2] SHELHAMER E, LONG J, DARRELL T. Fully convolu-tional networks for semantic segmentation[J/OL]. IEEETransactions on Pattern Analysis and Machine Intelligence,2024,39(4):640-651.

[3] RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[J]. Springer International Publishing,2015.

[4] CHEN L C, PAPANDREOU G, KOKKINOS I, et al. DeepLab: semantic image segmentation with deep convolu-tional nets, atrous convolution, and fully connected CRFs[J/OL]. IEEE Transactions on Pattern Analysis and MachineIntelligence,2023,40(4):834-848.

[5] 许伟鑫.基于深度卷积网络的肺结节分割与良恶性分类算法研究[D].重庆:重庆大学,2021.

[6] VASWANI A, SHAZEER N, PARMAR N, et al. Attention isall you need[C]//GUYON I, LUXBURG U, BENGIO S, et al. Advances in Neural Information Processing Systems,2022:45-48.

[7] CHEN J, LU Y, YU Q, et al. TransUNet: transformers makestrong encoders for medical image segmentation[J]. CornellUniversity - arXiv, Cornell University - arXiv,2021:123-128.

[8] ZHANG Y, LIU H, HU Q. TransFuse: fusing transformers andCNNs for medical image segmentation[C]//InternationalConference on Medical Image Computing and ComputerAssisted Intervention. Cham: Springer,2021:14-24

[9] GU A, DAO T. Mamba: linear-time sequence modeling withselective state spaces[J]. arXiv preprint arXiv,2024.

[10] ZHU L, LIAO B, ZHANG Q, et al. Vision Mamba: efficientvisual representation learning with bidirectional state spacemodel[J]. arXiv preprint arXiv,2024:34-37.