参考文献：Real-time target tracking of soft tissues in 3D ultrasound images based on robust visual information and mechanical simulation

期刊水平：MIA， medical imaging analysis

 图一：作者方法的计算流程图。深色的表示数据的输入和输出；浅色的表示作者的处理方法。

1. Introduction

Soft-tissue motion tracking is an active research area that consists in providing accurate evaluation about the location of anatomical structures. To do so, ultrasound imaging is often used since it is non-invasive, real-time and portable. Thus, several
ultrasound tracking approaches have been developed in order to estimate soft tissue displacements that are caused by physiological motions and manipulations by medical tools. These methods have gained significant interest for image-guided therapies such as radio-frequency ablation (RFA)  or high-intensity focused ultrasound (HIFU) (Pernot et al., 2004) that consist in eliminating tumors by delivering a local treatment on a targeted anatomical region. However, these tracking techniques remain sensitive to different ultrasound imaging shortcomings such as large ultrasound shadows, gain change and speckle noise. In this paper, we propose a novel tracking approach to tackle these limitations. Our method combines an intensity-based approach with a mechanical regularization（机械正规化）. We also propose an ultrasound-specific similarity criterion that has the advantage to be computationally efficient and robust to gain changes introduced by ultrasound imaging.

软组织的运动跟踪是一个活跃的研究领域，在于可以准确定位解剖结构位置。超声图像因为无损伤、实时、便携，被广泛使用。因此，一些研究人员设计了基于超声图像的跟踪方法，用于估计软组织的偏移（这些偏移主要是由生理运动和手术器械的操作造成的）。这些方法使得基于图像的治疗技术，如射频消融、高强度聚焦超声，受益良多（通过对靶向区域提供局部治疗进而消除肿瘤）。然而，这些方法仍然对超声图像的声影、斑点噪声、增益变化敏感，这篇文章，作者就是为了解决这些问题。作者的方法主要联合了基于灰度的方法和机械正则化。作者找到了一种适用于超声的相似性测度，使得相似度计算过程有效且鲁邦。

RFA参考文献：Higgins H, Berger D L. RFA for Liver Tumors: Does It Really Work?[J]. Oncologist, 2006, 11(7): 801-808.

HIFU参考文献：Pernot, M., Tanter, M., Fink, M., 2004. 3-D real-time motion correction in high-intensity focused ultrasound therapy. Ultrasound Med. Biol. 30 (9), 1239–1249.

2. Related work

2.1 基于特征的跟踪

基于表面和生物力学的匹配跟踪：Papademetris X, Sinusas A J, Dione D, et al. Estimation of 3-D left ventricular deformation from medical images using biomechanical models[J]. IEEE Transactions on Medical Imaging, 2002, 21(7): 786-800.

基于关键点的匹配跟踪（SIFT）：Schneider R J, Perrin D P, Vasilyev N V, et al. Real-time image-based rigid registration of three-dimensional ultrasound[J]. Medical Image Analysis, 2012, 16(2): 402-414.

为了提高对噪声的鲁棒性，这些方法可以基于贝叶斯框架来包含对目标形状的先验知识。

Angelova D S, Mihaylova L. Contour segmentation in 2D ultrasound medical images with particle filtering[J]. machine vision applications, 2011, 22(3): 551-561.

Rothlubbers, S., Schwaab, J., Jenne, J., Gunther, M., 2014. MICCAI CLUST 2014-Bayesian real-time liver feature ultrasound tracking. In: Proceedings of MICCAI Workshop on Challenge on Liver Ultrasound Tracking, p. 45

Zhang X, Gunther M, Bongers A, et al. Real-time organ tracking in ultrasound imaging using active contours and conditional density propagation[C]. international conference on medical imaging and augmented reality, 2010: 286-294.

但是，当关键特征由于声影等原因不可见时，这些方法就很容易跟踪失败。另一类方法是基于对强度损失函数的最小化；该损失函数使用单模相似度，如误差平方和（SSD）构建。

2.2 基于强度的损失函数优化

误差平方和 Sum of Squared Difference, SSD:

Lubke, D., Grozea, C., 2014. High performance online motion tracking in abdominal ultrasound imaging. In: Proceedings of MICCAI Workshop on Challenge on Liver Ultrasound Tracking, p. 29

Royer, L., Marchal, M., Le Bras, A., Dardenne, G., Krupa, A., 2015. Real-time tracking of deformable target in 3d ultrasound images. In: Proceedings of IEEE International Conference on Robotics and Automation

Yeung F, Levinson S F, Fu D, et al. Feature-adaptive motion tracking of ultrasound image sequences using a deformable mesh[J]. IEEE Transactions on Medical Imaging, 1998, 17(6): 945-956.

绝对误差和 Sum of Absolute Difference, SAD:

Touil B, Basarab A, Delachartre P, et al. Analysis of motion tracking in echocardiographic image sequences: Influence of system geometry and point-spread function[J]. Ultrasonics, 2010, 50(3):373-386.

交叉相关 Cross-Correlation, CC:

Basarab A, Liebgott H, Morestin F, et al. A method for vector displacement estimation with ultrasound imaging and its application for thyroid nodular disease[J]. Medical Image Analysis, 2008, 12(3):259-274.

De L V, Tschannen M, Székely G, et al. A Learning-Based Approach for Fast and Robust Vessel Tracking in Long Ultrasound Sequences[M]// Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013. Springer Berlin Heidelberg, 2013:518-25.

其实吧，如果目标的灰度不变，那么这些相似性测度都是很有效的，然而不幸的是，超声成像很特别，增益变化将会引起目标灰度强烈变化。

2.3 特异性的超声相似性测度

Baumann M, Mozer P, Daanen V, et al. Prostate biopsy tracking with deformation estimation[J]. Medical Image Analysis, 2012, 16(3):562-576. 提出了一种基于相关性的距离测度，这种测度还能够处理局部强度偏差（当超声束角度发生偏差就会出现灰度偏差）

Cohen B, Dinstein I. New maximum likelihood motion estimation schemes for noisy ultrasound images ☆[J]. Pattern Recognition, 2002, 35(2):455-463. 设计了新的相似度测度，该设计的灵感来源于超声的斑点噪声近似瑞利分布，所以对于超声斑点噪声污染严重的情况非常有用，但是对声影没有用。

Elen A, Choi H F, Loeckx D, et al. Three-dimensional cardiac strain estimation using spatio-temporal elastic registration of ultrasound images: a feasibility study[J]. IEEE Trans Med Imaging, 2008, 27(11):1580-1591. 建议采用互信息作为相似性测度，这样可以抵抗增益变化的影响，但是互信息计算量太大了。

Masum M A, Pickering M, Lambert A, et al. Accuracy assessment of Tri-plane B-mode ultrasound for non-invasive 3D kinematic analysis of knee joints[J]. Biomedical Engineering Online, 2014, 13(1):122. 建议使用条件方差和，虽然条件方差和可以降低对增益变化的敏感度，但是对于声影，条件方差和仍然束手无措。

2.4 Warping Model-变形模型

变形模型主要在于变形函数的设计。目前研究比较透得是平移变形函数 和 彷射变形函数； 研究的比较深的还是形变模型。

平移模型-平移形变函数：

Veronesi F, Corsi C, Caiani E G, et al. Nearly automated left ventricular long axis tracking on real time three-dimensional echocardiographic data[C]// Computers in Cardiology. IEEE, 2006:5-8.

彷射模型-彷射变形函数：

Wein, W., Cheng, J.-Z., Khamene, A., 2008. Ultrasound based respiratory motion compensation in the abdomen. In: Proceedings of MICCAI Worshop on Image Guidance and Computer Assistance for Soft tissue Interventions, 32, p. 294.

形变模型-形变变形函数：

A：基于传统的块匹配算法-计算两个连续帧之间小的图像块偏移量。

Basarab A, Liebgott H, Morestin F, et al. A method for vector displacement estimation with ultrasound imaging and its application for thyroid nodular disease[J]. Medical Image Analysis, 2008, 12(3):259-274.

De L V, Tschannen M, Székely G, et al. A Learning-Based Approach for Fast and Robust Vessel Tracking in Long Ultrasound Sequences[M]// Medical Image Computing and Computer-Assisted Intervention – MICCAI 2013. Springer Berlin Heidelberg, 2013:518-25.

Touil B, Basarab A, Delachartre P, et al. Analysis of motion tracking in echocardiographic image sequences: Influence of system geometry and point-spread function[J]. Ultrasonics, 2010, 50(3):373-386.

然而，块匹配方法不能表示高度表征局部的可变形的运动，因为它们假设在局部区域块中位移是刚性的。

B：稠密运动长场估计

为了解决1中遇到的问题，可以利用形变模型估计稠密运动场。典型的形变模型：

分段彷射模型：Royer, L., Marchal, M., Le Bras, A., Dardenne, G., Krupa, A., 2015. Real-time tracking
of deformable target in 3d ultrasound images. In: Proceedings of IEEE International Conference on Robotics and Automation.

薄板样条模型：Lee, D., Krupa, A., 2011. Intensity-based visual servoing for non-rigid motion compensation of soft tissue structures due to physiological motion using 4d ultrasound. In: Proceedings of IEEE International Conference on Intelligent Robots and Systems, pp. 2831–2836.

自由形变：Heyde, B., Claus, P., Jasaityte, R., Barbosa, D., Bouchez, S., Vandenheuvel, M., Wouters, P., Maes, F., Hooge, J.D., 2012. Motion and deformation estimation of cardiac ultrasound sequences using an anatomical B-spline transformation model. In: Proceedings of IEEE International Symposium on Biomedical Imaging, pp. 266–269.

为了确保鲁棒性，可以为这些方法添加时间-空间鲁棒性约束（或者是采用由粗到精的优化策略）。

时间-空间约束：Somphone, O., Allaire, S., Mory, B., Dufour, C., 2014. Live feature tracking in ultrasound liver sequences with sparse demons. In: Proceedings of MICCAI Workshop on Challenge on Liver Ultrasound Tracking, p. 53.、

由粗到精的优化策略：Mukherjee, R., Sprouse, C., Abraham, T., Hoffmann, B., McVeigh, E., Yuh, D., Burlina, P., 2011. Myocardial motion computation in 4d ultrasound. In: Proceedings of IEEE International Symposium on Biomedical Imaging. IEEE, pp. 1070–1073.

2.5 形变模型拓展研究

正反向配准方法提升跟踪精度:

Ledesma-Carbayo, M.J., Kybic, J., Desco, M., Santos, A., Unser, M., 2001. Cardiac motion analysis from ultrasound sequences using non-rigid registration. In: Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, pp. 889–896.

分组优化管理提升跟踪精度：

Metz, C., Klein, S., Schaap, M., van Walsum, T., Niessen, W., 2011. Nonrigid registration of dynamic medical imaging data using nD+t B-splines and a groupwise optimization approach. Med. Image Anal. 15 (2), 238–249.

Vijayan, S., Klein, S., Hofstad, E.F., Lindseth, F., Ystgaard, B., Lango, T., 2013. Validation of a non-rigid registration method for motion compensation in 4d ultrasound of the liver. In: Proceedings of IEEE International Symposium on Biomedical Imaging, pp. 792–795.

使用特定的网格代替标准化的矩形网格：

Heyde, B., Claus, P., Jasaityte, R., Barbosa, D., Bouchez, S., Vandenheuvel, M., Wouters, P., Maes, F., Hooge, J.D., 2012. Motion and deformation estimation of cardiac ultrasound sequences using an anatomical B-spline transformation model. In: Proceedings of IEEE International Symposium on Biomedical Imaging, pp. 266–269.

离群排斥（避免噪声干扰）：

Banerjee, J., Klink, C., Peters, E.D., Niessen, W.J., Moelker, A., van Walsum, T., 2015. Fast and robust 3d ultrasound registration block and game theoretic matching. Med. Image Anal. 20 (1), 173–183.

2.6 基于力学模型

除了这些仅基于视觉标准优化的技术之外，还提出了基于机械的跟踪方法，用于二维超声图像。

Loosvelt, M., Villard, P.-F., Berger, M.-O., 2014. Using a biomechanical model for tongue tracking in ultrasound images. In: Proceedings of IEEE Symp. on Biomedical Simulation. Springer, pp. 67–75.

Marami, B., Sirouspour, S., Fenster, A., W. Capson, D., 2014. Dynamic tracking of a deformable tissue based on 3d-2d MR-US image registration. Proceedings of SPIE Medical Imaging.

但是这种2D的力学模型无法解决out-of-plane运动。

为了解决这个问题，Yipeng Hu, Carter, T.J., Ahmed, H.U., Emberton, M., Allen, C., Hawkes, D.J., Barratt, D.C., 2011. Modelling prostate motion for data fusion during image-guided interventions. IEEE Trans. Med. Imaging 30 (11), 1887–1900.采用生物建模解决3D跟踪问题 (前列腺很简单，所以简单建模是可以的)。

然而，他们的方法需要在每个超声图像中手动识别前列腺的一些表面点，以驱动模型。总而言之，我们根据表1中的主要特性，提出了跟踪方法的分类。

据我们所知，目前还没有针对三维超声图像设计的实时跟踪方法，将鲁棒稠密运动估计和力学模型结合在一起。