因為超音波低成本及非侵入性與即時成像的特性，分析超音波影像序列之運動追蹤在臨床診斷中已經被廣泛的利用。在肌肉超音波影像中，由於軟組織形變所造成的非剛體運動往往增加運動估測的困難度。本論文中，針對二維肌肉超音波影像序列，我們提出了兩階段區塊匹配技術分析肌肉運動。這個方法是透過影像統計資料的選取及兩種大小不同匹配區塊的交替使用來提升影像追蹤準確度。使用紋理特徵與主成分兩種統計資料，能提供穩定的匹配兩區塊相似準則，甚至於運動較不穩定的範圍內，都能良好的確認與追蹤特徵顯著的區域。在第一階段，以較大的匹配區塊來掌握欲追蹤區域的全域資訊，並且挑選多個可能匹配的候選點。在第二階段，則在眾多候選點中以較小的匹配區塊比對欲追蹤區塊的區域變化，並找出最匹配的候選點。而我們提出了多樣板、次樣板及主成分分析等三種匹配準則供每個階段來使用。匹配時交替使用兩種大小不同的區塊，可在每一個運動追蹤的步驟中同時考慮到欲追蹤區域的全域資訊與區域變化。而從眾多候選點中挑選最佳候選點的兩階段式策略也可以提高追蹤的穩定度及準確度。再則，透過根據之前運動訊息來預測與修正的卡爾曼(Kalman)濾波器，可處理肌肉不穩定運動的問題。而在三維超音波影像中，同樣使用兩階段匹配。在第一個階段裡我們使用主成分分析做為匹配的準則，這是為了要克服因為低取樣率所造成的較大組織型變的問題。而第二階段，使用一半大小的樣板(而不是1/4大小)來做為匹配準則，主要是因為過小的樣板無法呈現欲追蹤區域的特徵。在實驗結果中，所提出的二維與三維超音波影像追蹤方法以活體肌肉骨骼超音波影像驗證，並與專家的評估比對。

Motion tracking by analyzing ultrasonic image sequences has become widespread in clinical diagnosis for its low cost, noninvasive, and real time imaging ability. Non-rigid motion in musculoskeletal ultrasonic image sequences increases the difficulty of motion estimation due to soft tissue deformation. According to the characters of 2D ultrasonic image sequences, two-staged block matching method through the selection of statistical information and the usage of alternate large and small matching block is proposed to estimate the muscular motion and can achieves high accuracy. Block matching technique based on the statistical information, such as texture features and principal components, can provide stable similarity criterion even on the ambiguity motion regions to determine and track the discrimination regions well. In the first stage, large matching block is used to handle the global information of matching pattern, and several matched candidates are chosen. In the second stage, the best candidate is selected from these candidates by comparing the local variation of matching pattern in small matching block. Three kinds of matching criteria, i.e. multi-template, sub-template, and principal component analysis, are developed and can be used in each matching stage. By using the large and small matching blocks alternatively, both global information and local variance can be taken into consideration simultaneously in the motion-tracking step. Two-staged strategy, which selects the best candidate from several candidates, can then obtain more accurate and stable tracking results. In addition, the Kalman prediction and correction based on previous motion information can be used to handle the problems of unstable movement. In 3D ultrasonic image sequence, two-staged block matching method is also used. In the first stage, the principal component analysis is used as the matching criterion in order to overcome the problem of large tissue deformation due to its low frame rate. In the second stage, the sub-template, with half (not quarter) of template size, is used as the matching template. It is because a too small template can not get the enough features to represent the matching region. In the experimental results, the accuracy is validated from in vivo musculoskeletal ultrasonic image sequences by comparing the results with the expert-defined ground truth.

[1] M. Suhling, et al., "Myocardial motion analysis from B-Mode echocardiograms," IEEE Transactions on Image Processing, vol. 14, 2005.
[2] B. D. Lucas and T. Kanade, "An iterative image registration technique with an application to stereo vision," presented at the International Joint Conference on Artificial Intelligence, 1981.
[3] F. Yeung, et al., "Multilevel and motion model-based ultrasonic speckle tracking algorithms," Ultrasound in Medicine & Biology, pp. 427-441, 1998.
[4] I. Pratikakis, et al., "Robust multi-scale non-rigid registration of 3D ultrasound images," Scale-Space and Morphology in Computer Vision, vol. 2106, pp. 389-397, 2001.
[5] B. Cohen and I. Dinstein, "New maximum likelihood motion estimation schemes for noisy ultrasound images," Pattern Recognition, vol. 35, pp. 455-463, 2002.
[6] C. I. Christodoulou, et al., "Texture-based classification of atherosclerotic carotid plaques," IEEE Transactions on Medical Imaging, vol. 22, pp. 902-913, 2003.
[7] C. H. Wu, et al., "Texture features for classification of ultrasonic liver images," IEEE Transactions on medical imaging, vol. 11, pp. 141-152, 1992.
[8] F. Yeung, et al., "Feature-adaptive motion tracking of ultrasound image sequences using a deformable mesh," IEEE Transactions on Medical Imaging, vol. 17, pp. 945-956, 1998.
[9] G. R. Bashford and O. T. von Ramm, "Ultrasound three-dimensional velocity measurements by feature tracking.pdf," IEEE Transaction on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 43, 1996.
[10] A. Krupa, et al., "Full motion tracking in ultrasound using image speckle information and visual servoing," presented at the IEEE International Conference on Robotics and Automation, 2007.
[11] W. M. Chen, et al., "3-D ultrasound texture classification using run difference matrix," Ultrasound in Medicine & Biology, vol. 31, pp. 763-770, 2005.
[12] Y. Zhan and D. Shen, "Automated segmentation of 3D US prostate images using statistical texture-based matching method," Medical Image Computing and Computer-Assisted Intervention - MICCAI 2003, vol. 2878, pp. 688-696, 2003.
[13] A. Myronenko, et al., "LV motion tracking from 3D echocardiography using textural and structural information," Medical Image Computing and Computer-Assisted Intervention – MICCAI 2007, vol. 4792, pp. 428-435, 2007.
[14] A. A. Morsy and O. T. von Ramm, "FLASH correlation: a new method for 3-D ultrasound tissue motiontracking and blood velocity estimation," IEEE Transaction on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 46, pp. 728-736, 1999.
[15] J. Kuo and O. T. von Ramm, "Three-dimensional motion measurements using feature tracking," IEEE Transaction on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 55, pp. 800-810, 2008.
[16] http://www.terason.com/products/t3000.asp.
[17] http://www.gehealthcare.com/euen/ultrasound/flash/voluson-e8/v2.html.
[18] J. C. Thompson and F. Netter, Netter's Concise Atlas of Orthopaedic Anatomy Icon Learning Systems LLC, 2002.
[19] E. N. Marieb and J. Mallatt, Human Anatomy, Benjamin Cummings, 2003.
[20] J. S. Lee, "Speckle analysis and smoothing of synthetic aperture radar images," Computer Graphics and Image Processing, vol. 17, pp. 24-32, 1981.
[21] C. P. Loizou, "Comparative evaluation of despeckle filtering in ultrasound imaging of the carotid artery," IEEE Transactions on Sonics and Ultrasonics, vol. 52, p. 1653, 2005.
[22] Y. Yu and S. T. Acton, "Speckle reducing anisotropic diffusion," IEEE Transactions on Image Processing, vol. 11, pp. 1260-1270, 2002.
[23] M. Amadasun, "Textural features corresponding to textural properties," IEEE Transactions on Systems, Man and Cybernetics, vol. 19, p. 1264, 1989.
[24] R. M. Haralick, "Textural features for image classification," IEEE Transactions on Systems, Man and Cybernetics, vol. 3, p. 610, 1973.
[25] G. H. Dunteman, Principal Components Analysis, SAGE, 1989.
[26] I. T. Jolliffe, Principal Component Analysis Springer, 2002.
[27] M. Turk, "Eigenfaces for recognition," Journal of Cognitive Neuroscience, vol. 3, p. 71, 1991.
[28] R. C. Gonzalez and R. E. Woods, Digital Image Processing, Prentice Hall, 2002.
[29] R. E. Kalman, "A new approach to linear filtering and prediction problems," Transactions of the ASME-Journal of basic Engineering, pp. 35-45, 1960.
[30] G. Welch and G. Bishop, An introduction to the Kalman filter, Department of Computer Science, Universe of North Carolina at Chapel Hill, 2006.
[31] M. Zaki, "EMBOT: An enhanced motion-based object tracker," The Journal of Systems and Software, vol. 69, p. 149, 2004.
[32] A. H. Mohamed, "Adaptive Kalman filtering for INS/GPS," Journal of Geodesy, vol. 73, p. 193, 1999.