北京生物医学工程

基于Transformer的血管内超声图像分割

Transformer-based intravascular ultrasound image segmentation

作者：李佳松曹洪帅舒丽霞蔺嫦燕

单位：首都医科大学附属北京安贞医院(北京 100029)   首都医科大学临床生物力学应用基础研究所北京市重点实验室（北京 100069）<br />通讯作者：蔺嫦燕。E-mail: llbl@sina.com

分类号：R318.04

出版年·卷·期（页码）：2023·42·1（16-20）

摘要：

目的提出一种基于Transformer的血管内超声图像分割方法，以解决冠状动脉钙化病变血管内超声图像显影不完全导致的分割管腔、外弹力膜和钙化斑块精度不高的问题。方法采用深度学习方法，在UNet结构的基础上用多分辨率卷积层提取不同大小类别特征，在特征编码模块与特征解码模块之间使用Transformer联系上下文信息，同时分割管腔、外弹力膜和钙化斑块。最后以34个40 MHz血管内超声序列得到的训练集720张和测试集240张为例对上述方法进行训练和测试。结果外弹力膜分割杰卡德系数 (Jaccard index，JI)为0.92，豪斯多夫距离（Hausdorff distance，HD)为0.84 mm；管腔分割JI为0.85， HD为1.44 mm；钙化分割JI为0.67， HD为0.68 mm。结论该方法能够提升血管内超声图像的分割精度，并且在钙化病变血管显影不完全时能够保持分割效果。

Objective To propose a intravascular ultrasound image segmentation method based on Transformer，and to solve the problem of low precision in segmenting of lumen, external elastic membrane and calcified plaque caused by incomplete development of intravascular ultrasound image of coronary artery calcification disease. Methods A deep learning technique was employed to extract different size category features using multi-resolution convolutional layers based on UNet structure. Additionally, Transformer was used to link contextual information between feature encoding module and feature decoding module. For training and testing the above method, 34 40 MHz intravascular ultrasound sequences were captured, along with 720 training sets and 240 test sets. Results External elastic membrane segmentation JI was 0.92，HD was 0.84 mm；lumen segmentation JI was 0.85，HD was 1.44 mm；calcification segmentation JI was 0.67，HD was 0.68 mm. Conclusions The method can improve the segmentation accuracy of intravascular ultrasound images and maintain the segmentation effect when the development of calcified lesion vessels is incomplete.

参考文献：

[1] Nissen SE, Yock P. Intravascular ultrasoundnovel pathophysiological insights and current clinical applications [J]. Circulation, 2001, 103: 604-16.
[2] Wang X, Matsumura M, Mintz GS, et al. In vivo calcium detection by comparing optical coherence tomography, intravascular ultrasound, and angiography [J]. JACC Cardiovasc Imaging, 2017, 10(8): 869-79.
[3] 王伟民, 霍勇, 葛均波. 冠状动脉钙化病变诊治中国专家共识（2021版）[J]. 中国介入心脏病学杂志, 2021,29（5）：251-259.
[4] Huang C, Wang J, Xie Q, et al. Analysis methods of coronary artery intravascular images: A review [J]. Neurocomputing, 2022 ，489: 27-39.
[5] Jodas DS, Pereira AS, Tavares JMR. Automatic segmentation of the lumen region in intravascular images of the coronary artery [J]. Medical image analysis, 2017, 40: 60-79.
[6] Hammouche A, Cloutier G, Tardif JC, et al. Automatic IVUS lumen segmentation using a 3D adaptive helix model [J]. Computers in Biology and Medicine, 2019, 107: 58-72.
[7] Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical image computing and computer-assisted intervention. Singapore, Singapore: Springer, Cham, 2015: 234-241.
[8] Yang J, Faraji M, Basu A. Robust segmentation of arterial walls in intravascular ultrasound images using Dual Path U-Net [J]. Ultrasonics, 2019, 96: 24-33.
[9] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need [J]. Advances in neural information processing systems, 2017, 30: 5998-6008.
[10] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16x16 words: Transformers for image recognition at scale [EB/OL]. arXiv .2020-10-22[2021-06-03]. https://doi.org/10.48550/arXiv.2010.11929.
[11] Chen J, Lu Y, Yu Q, et al. Transunet: Transformers make strong encoders for medical image segmentation [EB/OL]. arXiv.2021-02-08[2021-12-09].https://doi.org/10.48550/arXiv.2102.04306
[12] Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]//International conference on machine learning. Lille, France: PMLR, 2015: 448-456.
[13] Glorot X, Bordes A, Bengio Y. Deep sparse rectifier neural networks[C]//Proceedings of the fourteenth international conference on artificial intelligence and statistics. Ft. Lauderdale, FL, USA: JMLR Workshop and Conference Proceedings, 2011: 315-323.
[14] Lin A, Chen B, Xu J, et al. Ds-transunet: Dual swin transformer u-net for medical image segmentation [J]. IEEE Transactions on Instrumentation and Measurement, 2022,71：4005615.
[15] Yan X, Tang H, Sun S, et al. After-unet: Axial fusion transformer unet for medical image segmentation[C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa, HI, USA: IEEE press, 2022: 3971-3981.
[16] Liu Z, Lin Y, Cao Y, et al. Swin transformer: Hierarchical vision transformer using shifted windows[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Montreal, QC, Canada: IEEE press, 2021: 10012-10022.
[17] 汪琳琳, 施俊, 韩振奇,等. 结合卷积神经网络与图卷积网络的乳腺癌病理图像分类研究[J]. 北京生物医学工程, 2021, 40(2): 130-138.
Wang LL，Shi J，Han ZQ，et al. Research on breast cancer pathological image classification combined with convolution neural network and graph convolution network[J]. Beijing Biomedical Engineering, 2021, 40(2): 130-138.
[18] Eigen D, Fergus R. Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture[C]//Proceedings of the IEEE international conference on computer vision. Washington, DC, USA: IEEE press, 2015: 2650-2658.
[19] Shen W, Zhou M, Yang F, et al. Multi-scale convolutional neural networks for lung nodule classification[C]//International conference on information processing in medical imaging. Sabhal Mor Ostaig, Isle of Skye, United Kingdom: Springer, Cham, 2015: 588-599.
[20] Madhavan MV, Tarigopula M, Mintz GS, et al. Coronary artery calcification: pathogenesis and prognostic implications [J]. Journal of The American College of Cardiology, 2014, 63(17): 1703-1714.
[21] Wu H, Chen S, Chen G, et al. FAT-Net: Feature adaptive transformers for automated skin lesion segmentation [J]. Medical Image Analysis, 2022, 76: 102327.

服务与反馈：

【文章下载】【加入收藏】

提示：您还未登录，请登录！点此登录