北京生物医学工程

基于三维卷积神经网络的颈部ＴＯＦＭＲＡ图像的血管自动分割

Automatic vessel segmentation of neck TOF MRA image based on 3D convolutional neural network

作者：邱伟陈硕魏寒宇李睿

单位：清华大学医学院生物医学工程系生物医学影像研究中心（北京 100084）。 <p>通信作者：李睿，博士，研究员。E-mail：leerui@tsinghua.edu.cn</p>

关键词：动脉粥样硬化；卷积神经网络；时间飞跃法磁共振血管造影图像；三维切块；血管分割

分类号：R318.04

出版年·卷·期（页码）：2022·41·6（551-557）

摘要：

目的提出一种基于三维卷积神经网络（convolutional neural network, CNN）的分割模型，通过设计网络结构和调整参数，实现对颈部时间飞跃法磁共振血管造影（time of flight magnetic resonance angiography, TOF MRA）图像的自动精准血管分割。方法三维颈部TOF MRA图像数据来源于一项无症状老年人心脑血管病发病风险研究中的166例受试者。首先由血管影像临床经验的放射诊断医生运用Mimics软件对颈部动脉血管进行手动标注，再按照8:1:1的比例将数据随机分为训练集（132例）、验证集（17例）和测试集（17例）。针对TOF MRA图像稀疏性和高对比度的特点，采用3D CNN的优化模型，通过在U-Net网络的编码和解码路径中多层级的加入专门的模块和不同分辨率的原始图像，更好地学习和利用到图像的独有特征。为研究输入图像的大小对网络性能的影响，从原始图像中分别裁剪出3种不同尺寸大小的三维切块来进行模型训练。在十折交叉验证下，采用Dice系数的平均值和标准差，以及灵敏度和特异度对模型的分割性能进行评价。采用单因素方差分析对比3种切块尺寸下的测试实验。结果所提出的新模型取得了最高的分割Dice值（0.932 0）、灵敏度（0.918 6）和特异度（0.999 6），以及最小的Dice值标准差（0.005 1）。ANOVA显著性水平为，表明了不同切块尺寸下的模型分割结果有显著性差异，尺寸增大，模型分割结果更好。结论所提出的基于3D CNN的优化模型在TOF MRA图像血管自动精准分割上优于已有方法。此外，增加模型输入图像的尺寸大小有助于提高分割性能。

Objective Propose an automatic segmentation model based on 3D convolutional neural network to segment neck vessel from the time-of-flight magnetic resonance angiography(TOF MRA) data. Methods A total of 166 neck TOF MRA images were selected from the Cardiovascular Risk of Old Population(CROP) study. Vessel segmentation labels were manually delineated by vascular imaging professionals in 3D TOF images (Materialise Mimics, Mimics Medical 17.0), and were subsequently examined by experienced imaging experts. Randomly, 132 images were assigned into training dataset, 17 images were assigned into validating dataset, 17 images were assigned into test dataset according to a ratio of 8∶1∶1. Based on the sparsity and high contrast of TOF MRA image, a 3D CNN model suitable for TOF MRA data were proposed by adding special modules and images to the encoding and decoding path of U-Net network. The unique features of the image can be better learned and utilized. In addition, in order to study the influence of the size of the input on the network performance, three different sizes of 3D patch randomly cropped from the original image were selected for model training and predicting. The average Dice coefficients, sensitivity, specificity and standard deviation of the Dice coefficients under 10-fold cross-validation were used to evaluate the segmentation performance of the model. In this comparative experiments, One-way ANOVA was carried out for the experiments of the three kinds of patch size. Results Proposed 3D CNN model obtained the highest segmentation Dice coefficient value(0.932 0), sensitivity(0.918 6) and specificity(0.999 6), and the smallest standard deviation (0.005 1). In the results of One-way ANOVA, the p<0.01, indicated that the results of different patch sizes had significant differences. And as the size increased, the segmentation result of the model was better. Conclusions The experimental results showed that the proposed model can automatically segment the neck vascular from TOF-MRA volumes and outperformed the state of the art. Besides, increasing the size of the input image can improve the segmentation performance.

参考文献：

[1] Wu S, Wu B, Liu M, et al. Stroke in China: advances and challenges in epidemiology, prevention, and management[J]. The Lancet Neurology, 2019,18(4): 394-405.
[2] Pega F, Náfrádi B, Momen NC, et al. Global, regional, and national burdens of ischemic heart disease and stroke attributable to exposure to long working hours for 194 countries, 2000-2016: a systematic analysis from the WHO/ILO Joint Estimates of the Work-related Burden of Disease and Injury[J]. Environment International, 2021,154: 106595.
[3] Bos D, Arshi B, van den Bouwhuijsen Q, et al. Atherosclerotic Carotid Plaque Composition and Incident Stroke and Coronary Events[J]. Journal of the American College of Cardiology, 2021, 77(11):1426-1435.
[4] 王拥军,李子孝,谷鸿秋,等.中国卒中报告2019(中文版)(1)[J].中国卒中杂志,2020,15(10):1037-1043.
[5] Sun Y, Shi YM, Xu P. The clinical research progress of vertebral artery dominance and posterior circulation ischemic stroke[J]. Cerebrovascular Diseases (Basel, Switzerland), 2022, 51(5): 553-556.?
[6] Debrey SM, Yu H, Lynch JK, et al. Diagnostic accuracy of magnetic resonance angiography for internal carotid artery disease: a systematic review and meta-analysis[J]. Stroke, 2008, 39(8):2237-2248.?
[7] Munio M, Darcourt J, Gollion C, et al. Large artery intracranial stenosis in young adults with ischaemic stroke[J]. Revue Neurologique, 2021,178(3):206-212.?
[8] Villablanca JP, Nael K, Habibi R, et al. 3 T contrast-enhanced magnetic resonance angiography for evaluation of the intracranial arteries: comparison with time-of-flight magnetic resonance angiography and multislice computed tomography angiography[J]. Investigative Radiology, 2006,41(11): 799-805.
[9] Josephson CB, White PM, Krishan A, et al. Computed tomography angiography or magnetic resonance angiography for detection of intracranial vascular malformations in patients with intracerebral haemorrhage[J]. The Cochrane Database of Systematic Reviews, 2014(9): CD009372.
[10] Wang X, Benson J, Jagadeesan B, et al. Giant cerebral aneurysms: comparing CTA, MRA, and digital subtraction angiography assessments[J]. Journal of Neuroimaging, 2020,30(3): 335-341.?
[11] 张琨.核磁共振血管成像与螺旋CT血管成像技术诊断脑血管疾病的价值分析[J].中国医疗器械信息,2020,26(13): 65-66.
Zhang K. Analysis of the value of nuclear magnetic resonance angiography and spiral CT angiography in the diagnosis of cerebrovascular diseases[J]. China Medical Device Information,2020,26(13):65-66.
[12] 李继凡,陈硕,章强,等. 基于U?Net神经网络的多模态MR颈动脉血管成像的分割方法研究[J].中华放射学杂志, 2019,53(12):1091-1095.
Li JF, Chen S, Zhang Q, et al. The study on the segmentation of carotid vessel wall in multicontrast MR images based on U?Net neural network[J]. Chinese Journal of Radiology,2019,53(12):1091-1095.
[13] Settecase F, Rayz VL. Advanced vascular imaging techniques[J]. Handbook of Clinical Neurology, 2021,176: 81-105.?
[14] Osmanodja F, Scheitz JF, Fiebach JB, et al. Can intracranial time-of-flight-MR angiography predict extracranial carotid artery stenosis?[J]. Journal of Neurology, 2021,269:2743-2749.?
[15] Wilson DL, Noble JA. Segmentation of cerebral vessels and aneurysms from MR angiography data[C]// International Conference on Information Processing in Medical Imaging. Berlin: Springer, Berlin, Heidelberg, 1997: 423-428.
[16] Xiao R, Ding H, Zhai F, et al. Cerebrovascular segmentation of TOF-MRA based on seed point detection and multiple-feature fusion[J]. Computerized Medical Imaging and Graphics, 2018, 69:1-8.?
[17] 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. Berlin: Springer, Cham, 2015: 234-241.?
[18] Cicek ?, Ahmed Abdulkadir, Lienkamp SS, et al. 3D u-net: learning dense volumetric segmentation from sparse annotation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, Cham, 2016: 424–432.?
[19] Litjens G, Kooi T, Bejnordi BE, et al. A survey on deep learning in medical image analysis[J]. Medical Image Analysis, 2017,42: 60?88.?
[20] Norman B, Pedoia V, Majumdar S. Use of 2D U?Net convolutional neural networks for automated cartilage and meniscus segmentation of knee MR imaging data to determine relaxometry and morphometry[J]. Radiology, 2018, 288(1): 177?185.
[21] Seo H, Badiei Khuzani M, Vasudevan V, et al. Machine learning techniques for biomedical image segmentation: An overview of technical aspects and introduction to state-of-art applications[J]. Medical Physics, 2020,47(5): e148-e167.?
[22] Phellan R, Peixinho A, Falc?o A, et al. Vascular segmentation in TOF MRA images of the brain using a deep convolutional neural network[C]// Intravascular Imaging and Computer Assisted Stenting, and Large-Scale Annotation of Biomedical Data and Expert Label Synthesis. Berlin: Springer, Cham: 2017: 39-46.?
[23] Livne M, Rieger J, Aydin OU, et al. A U-net deep learning framework for high performance vessel segmentation in patients with cerebrovascular disease[J]. Frontiers in Neuroscience, 2019, 13: 97.?
[24] Sanches P, Meyer C, Vigon V, et al. Cerebrovascular network segmentation of MRA images with deep learning[C]// 2019 IEEE 16th International Symposium on Biomedical Imaging(ISBI 2019). Venice, Italy: IEEE Press, 2019: 768-771.
[25] Zhu C, Wang X, Teng Z, et al. Cascaded residual U-net for fully automatic segmentation of 3D carotid artery in high-resolution multi-contrast MR images[J]. Physics in Medicine and Biology, 2021,66(4): 045033.
[26] Han Y, Guan M, Zhu Z, et al. Assessment of longitudinal distribution of subclinical atherosclerosis in femoral arteries by three-dimensional cardiovascular magnetic resonance vessel wall imaging[J]. Journal of Cardiovascular Magnetic Resonance, 2018, 20: 60.?
[27] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Las Vegas, NV, USA: IEEE Press, 2016: 770-778.
[28] Hu J, Shen L, Sun G, et al. Squeeze-and-excitation networks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA: IEEE Press, 2018: 7132-7141.?

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