双域多尺度状态空间网络下的口腔颌面全景X射线图像分割算法研究

李冰; 胡伟杰; 刘侠

doi:10.11999/JEIT250639

双域多尺度状态空间网络下的口腔颌面全景X射线图像分割算法研究

doi: 10.11999/JEIT250639 cstr: 32379.14.JEIT250639

李冰^{1, 2},
胡伟杰^{1, 2, ,},
刘侠^{1, 2}

1.
哈尔滨理工大学自动化学院哈尔滨 150080
2.
黑龙江省复杂智能系统与集成重点实验室哈尔滨 150080

基金项目: 国家自然科学基金面上项目(61672197)，黑龙江省自然科学基金项目(LH00F035)

详细信息

作者简介:
李冰：女，博士，教授，研究方向为医学图像处理与分析，网络控制系统分析与综合

胡伟杰：女，硕士生，研究方向为医学图像处理与分析、模式识别

刘侠：男，博士，教授，研究方向为医学图像处理、模式识别、机器学习

通讯作者:
胡伟杰　huweijie0917@163.com

中图分类号: TP391.4; TP751; TP317.4
计量
- 文章访问数: 24
- HTML全文浏览量: 11
- PDF下载量: 4
- 被引次数: 0
出版历程
- 修回日期: 2025-12-22
- 录用日期: 2025-12-22
- 网络出版日期: 2025-12-29

Research on Segmentation Algorithm of Oral and Maxillofacial Panoramic X-ray Images under Dual-domain Multiscale State Space Network

LI Bing^{1, 2},
HU Weijie^{1, 2
, ,},
LIU Xia^{1, 2}

1.
School of Automation, Harbin University of Science and Technology, Harbin 150080, China
2.
Heilongjiang Provincial Key Laboratory of Complex Intelligent System and Integration, Harbin 150080, China

Funds: The National Natural Science Foundation of China (61172167), Heilongjiang Provincial Natural Science Foundation Project (LH00F035)

摘要

摘要: 针对口腔颌面全景X射线图像中存在的形态变异显著、牙体－牙龈边界模糊以及牙周组织灰度值重叠等问题，该研究提出基于双域多尺度状态空间网络的口腔颌面全景Ｘ射线图像分割算法。空间域利用视觉状态空间块建立牙弓动态传播模型，并利用微分方程实现跨象限长程关联捕捉。特征域构建可变形多尺度注意力金字塔，并利用通道－空间注意力动态加权关键解剖标志的灰度渐变特征，解析牙体－牙龈模糊边界。双域特征进一步通过三重注意力融合机制，强化解剖标注的语义表达。实验表明，该算法在颌面全景Ｘ射线图像分割任务中取得显著效果，戴斯系数(Dice)达93.8%，豪斯多夫距离(HD95)为18.73像素，充分验证了算法的有效性。
- 口腔颌面全景X射线图像分割 /
- 视觉状态空间块 /
- 可变形多尺度注意力金字塔 /
- 三重注意力融合
Abstract: Objective To address significant morphological variations, blurred boundaries between teeth and gums, and overlapping gray levels in periodontal tissues within panoramic X-ray images of the oral and maxillofacial region, this paper introduces a state-space model based on Mamba, a recent neural network architecture. This model retains the advantage of Convolutional Neural Networks (CNNs) in extracting local image features while avoiding the high computational complexity of Transformers. Building on this, the study proposes a dual-domain multiscale state space network-based segmentation algorithm for panoramic X-ray images of the oral and maxillofacial region, ultimately achieving significant improvements in segmentation accuracy and model efficiency. Methods This model adopts a network architecture primarily based on encoding-decoding, comprising a dual-branch encoder for acquiring global and local information, a decoder for feature restoration, and skip connections for conveying fused encoding path feature maps. During the decoding phase, fused features gradually recover resolution and reduce channel count through deconvolution combined with sampling modules, ultimately outputting a 2-channel segmentation map. Results and Discussions This study validated the contribution of each module to model performance through ablation experiments, as shown in Table 1. This model achieved significant performance improvements: in segmentation accuracy, the Dice score increased by 5.69 percentage points to 93.86%, while the HD95 value decreased by 2.97 mm to 18.73 mm, achieving 94.57% accuracy. Regarding model efficiency, the model has a size of only 81.23 MB with 90.1 million parameters, which is significantly lower than that of the baseline model, achieving synergistic optimization of accuracy improvement and model compression. When compared with seven mainstream medical image segmentation models under identical experimental conditions, as shown in Table 2, DMSS-Net achieves higher segmentation accuracy while maintaining a model size comparable to or smaller than that of similarly scaled Transformer models. Conclusions This paper proposes a dual-domain multiscale state-space network-based segmentation algorithm for panoramic X-ray images of the oral and maxillofacial region. The algorithm centers on a dual-domain fusion network framework. This framework significantly enhances the model’s ability to capture long-range dependencies in dental images and enables precise segmentation of blurred boundaries. An innovative spatial domain design helps address the challenge of capturing long-range context amid the dynamic dental arch morphology. In addition, a mechanism that enhances the feature domain not only improves the detection of low-contrast features but also increases robustness against interference.
- Oral and maxillofacial panoramic X-ray image segmentation /
- Visual state space blocks /
- Deformable multi-scale attention pyramids /
- Triple attention fusion

HTML全文

图 1 双域多尺度状态空间网络

下载: 全尺寸图片幻灯片

图 3 选择性扫描模块工作机制

下载: 全尺寸图片幻灯片

图 2 视觉状态空间块

下载: 全尺寸图片幻灯片

图 4 可变形多尺度注意力金字塔

下载: 全尺寸图片幻灯片

图 5 空间与通道注意力协同模块

下载: 全尺寸图片幻灯片

图 6 三重注意力融合机制

下载: 全尺寸图片幻灯片

图 7 口腔颌面全景X射线数据集

下载: 全尺寸图片幻灯片

图 8 消融实验结果图

下载: 全尺寸图片幻灯片

图 9 常规场景与挑战性场景下的分割效果

下载: 全尺寸图片幻灯片

图 10 各网络分割效果对比

下载: 全尺寸图片幻灯片

图 11 各网络指标对比

下载: 全尺寸图片幻灯片

表 1 消融实验指标对比

	Dice(%)	HD95	Accuracy(%)	Recall(%)	Precision(%)	模型大小/MB	参数量/M
baseline	88.17	21.70	93.03	87.97	86.73	84.24	93.19
baseline_A	89.41	22.01	93.67	88.24	87.07	79.23	88.43
baseline_B	89.16	22.10	93.74	88.32	87.42	81.15	89.16
baseline_C	89.73	21.53	93.86	88.87	87.13	78.59	85.63
baseline_AB	91.32	20.71	94.07	89.03	87.26	82.15	91.52
baseline_AC	91.24	19.82	94.20	89.24	87.34	83.41	92.31
baseline_BC	91.53	19.77	94.13	89.37	87.42	81.68	90.74
Ours	93.86	18.73	94.57	90.46	88.03	81.23	90.10

下载: 导出CSV

表 2 对比实验指标对比

模型名称	Dice (%)	HD95 (mm)	Accuracy (%)	Recall (%)	Precision (%)	模型大小 (MB)	参数量 (M)	GFlops (G)	FPS
TransUnet	91.43	24.13	92.02	88.76	86.31	82.14	88.52	29.34	18.24
Swin-Unet	89.03	24.63	92.63	89.13	86.81	84.61	92.31	40.93	11.47
VM-Unet	89.63	23.07	93.08	89.47	87.23	83.15	90.74	6.87	34.71
DA-TransUnet	91.25	20.29	93.24	89.73	87.78	87.78	80.63	31.42	16.87
Ege-Unet	88.18	32.64	92.57	86.93	86.04	52.87	61.91	13.73	27.38
TinyU-Net	86.27	35.83	91.74	86.02	85.94	48.65	58.31	4.18	45.63
SAMUS	88.51	31.83	95.02	88.83	86. 57	162.17	183.19	51.62	4.53
0urs	93.86	18.73	94.30	90.46	88.03	81.23	90.10	24.93	22.14

下载: 导出CSV

参考文献(27)

[1]	RUAN Jiacheng, XIE Mingye, GAO Jingsheng, et al. EGE-UNet: An efficient group enhanced UNet for skin lesion segmentation[C]. Proceedings of the 26th International Conference on Medical Image Computing and Computer Assisted Intervention, Vancouver, Canada: Springer, 2023: 481–490. doi: 10.1007/978-3-031-43901-8_46.
[2]	CHEN Junren, CHEN Rui, WANG Wei, et al. TinyU-Net: Lighter yet better U-Net with cascaded multi-receptive fields[C]. Proceedings of the 27th International Conference on Medical Image Computing and Computer Assisted Intervention, Marrakesh, Morocco: Springer, 2024: 626–635. doi: 10.1007/978-3-031-72114-4_60.
[3]	CAO Hu, WANG Yueyue, CHEN J, et al. Swin-Unet: Unet-like pure transformer for medical image segmentation[C]. Proceedings of the 17th European Conference on Computer Vision, Tel Aviv, Israel: Springer, 2023: 205–218. doi: 10.1007/978-3-031-25066-8_9.
[4]	CHEN Jieneng, LU Yongyi, YU Qihang, et al. TransUNet: Transformers make strong encoders for medical image segmentation[J]. arXiv preprint arXiv: 2102.04306, 2021. doi: 10.48550/arXiv.2102.04306. (查阅网上资料,不确定本条文献的格式和类型,请确认).
[5]	SUN Guanqun, PAN Yizhi, KONG Weikun, et al. DA-TransUNet: Integrating spatial and channel dual attention with transformer U-Net for medical image segmentation[J]. Frontiers in Bioengineering and Biotechnology, 2024, 12: 1398237. doi: 10.3389/fbioe.2024.1398237.
[6]	LEE H H, BAO Shunxing, HUO Yuankai, et al. 3D UX-Net: A large kernel volumetric convnet modernizing hierarchical transformer for medical image segmentation[C]. Proceedings of the 11th International Conference on Learning Representations, Kigali, Rwanda: ICLR, 2023.
[7]	ZHOU Hongyu, GUO Jiansen, ZHANG Yinghao, et al. nnFormer: Volumetric medical image segmentation via a 3D transformer[J]. IEEE Transactions on Image Processing, 2023, 32: 4036–4045. doi: 10.1109/TIP.2023.3293771.
[8]	HATAMIZADEH A, NATH V, TANG Yucheng, et al. Swin UNETR: Swin transformers for semantic segmentation of brain tumors in MRI images[C]. Proceedings of the 7th International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries, Springer, 2021: 272–284. doi: 10.1007/978-3-031-08999-2_22. (查阅网上资料,未找到出版地信息,请确认).
[9]	GU A and DAO T. Mamba: Linear-time sequence modeling with selective state spaces[J]. arXiv preprint arXiv: 2312.00752, 2023. doi: 10.48550/arXiv.2312.00752. (查阅网上资料,不确定本条文献的格式和类型,请确认).
[10]	RUAN Jiacheng, LI Jincheng, and XIANG Suncheng. VM-UNet: Vision mamba UNet for medical image segmentation[J]. ACM Transactions on Multimedia Computing, Communications and Applications, 2025. doi: 10.1145/3767748. (查阅网上资料,未找到卷期页码信息,请确认).
[11]	HAO Jing, ZHU Yonghui, HE Lei, et al. T-Mamba: A unified framework with long-range dependency in dual-domain for 2D & 3D tooth segmentation[J]. arXiv preprint arXiv: 2404.01065, 2024. doi: 10.48550/arXiv.2404.01065. (查阅网上资料,不确定本条文献的格式和类型,请确认).
[12]	LIN Xian, XIANG Yangyang, YU Li, et al. Beyond adapting SAM: Towards end-to-end ultrasound image segmentation via auto prompting[C]. Proceedings of the 27th International Conference on Medical Image Computing and Computer Assisted Intervention, Marrakesh, Morocco: Springer, 2024: 24–34. doi: 10.1007/978-3-031-72111-3_3.
[13]	LIN P L, HUANG P Y, HUANG P W, et al. Teeth segmentation of dental periapical radiographs based on local singularity analysis[J]. Computer Methods and Programs in Biomedicine, 2014, 113(2): 433–445. doi: 10.1016/j.cmpb.2013.10.015.
[14]	MAHDI F P and KOBASHI S. Quantum particle swarm optimization for multilevel thresholding-based image segmentation on dental X-ray images[C]. Proceedings of the Joint 10th International Conference on Soft Computing and Intelligent Systems and 19th International Symposium on Advanced Intelligent Systems, Toyama, Japan: IEEE, 2018: 1148–1153. doi: 10.1109/SCIS-ISIS.2018.00181.
[15]	SON L H and TUAN T M. A cooperative semi-supervised fuzzy clustering framework for dental X-ray image segmentation[J]. Expert Systems with Applications, 2016, 46: 380–393. doi: 10.1016/j.eswa.2015.11.001.
[16]	PUSHPARAJ V, GURUNATHAN U, ARUMUGAM B, et al. An effective numbering and classification system for dental panoramic radiographs[C]. Proceedings of the 4th International Conference on Computing, Communications and Networking Technologies, Tiruchengode, India: IEEE, 2013: 1–8. doi: 10.1109/ICCCNT.2013.6726480.
[17]	ALSMADI M K. A hybrid Fuzzy C-Means and Neutrosophic for jaw lesions segmentation[J]. Ain Shams Engineering Journal, 2018, 9(4): 697–706. doi: 10.1016/j.asej.2016.03.016.
[18]	KOCH T L, PERSLEV M, IGEL C, et al. Accurate segmentation of dental panoramic radiographs with U-NETS[C]. Proceedings of the 2019 IEEE 16th International Symposium on Biomedical Imaging, Venice, Italy: IEEE, 2019: 15–19. doi: 10.1109/ISBI.2019.8759563.
[19]	ZANNAH R, BASHAR M, MUSHFIQ R B, et al. Semantic segmentation on panoramic dental X-ray images using U-Net architectures[J]. IEEE Access, 2024, 12: 44598–44612. doi: 10.1109/ACCESS.2024.3380027.
[20]	IMAK A, ÇELEBI A, POLAT O, et al. ResMIBCU-Net: An encoder–decoder network with residual blocks, modified inverted residual block, and bi-directional ConvLSTM for impacted tooth segmentation in panoramic X-ray images[J]. Oral Radiology, 2023, 39(4): 614–628. doi: 10.1007/s11282-023-00677-8.
[21]	LI Yunxiang, WANG Shuai, WANG Jun, et al. GT U-Net: A U-net like group transformer network for tooth root segmentation[C]. Proceedings of the Machine Learning in Medical Imaging: 12th International Workshop, Strasbourg, France: Springer, 2021: 386–395. doi: 10.1007/978-3-030-87589-3_40.
[22]	SHENG Chen, WANG Lin, HUANG Zhenhuan, et al. Transformer-based deep learning network for tooth segmentation on panoramic radiographs[J]. Journal of Systems Science and Complexity, 2023, 36(1): 257–272. doi: 10.1007/s11424-022-2057-9.
[23]	LI Pengcheng, GAO Chenqiang, LIAN Chunfeng, et al. Spatial prior-guided bi-directional cross-attention transformers for tooth instance segmentation[J]. IEEE Transactions on Medical Imaging, 2024, 43(11): 3936–3948. doi: 10.1109/TMI.2024.3406015.
[24]	LIU Yue, TIAN Yunjie, ZHAO Yuzhong, et al. VMamba: Visual state space model[C]. Proceedings of the 38th International Conference on Neural Information Processing Systems, Vancouver, Canada: Curran Associates Inc. , 2024: 3273.
[25]	HOWARD A G, ZHU Menglong, CHEN Bo, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications[J]. arXiv preprint arXiv: 1704.04861, 2017. doi: 10.48550/arXiv.1704.04861. (查阅网上资料,不确定本条文献的格式和类型,请确认).
[26]	SI Yunzhong, XU Huiying, ZHU Xinzhong, et al. SCSA: Exploring the synergistic effects between spatial and channel attention[J]. Neurocomputing, 2025, 634: 129866. doi: 10.1016/j.neucom.2025.129866.
[27]	SUN Hongkun, XU Jing, and DUAN Yuping. ParaTransCNN: Parallelized transcnn encoder for medical image segmentation[J]. arXiv preprint arXiv: 2401.15307, 2024. doi: 10.48550/arXiv.2401.15307. (查阅网上资料,不确定本条文献的格式和类型,请确认).