本論文的主要目的是建立一套針對二維X光膝關節前後照影像偵測出膝關節骨刺所在位置與範圍的系統，提供醫師一個客觀的依據，讓醫師能夠在臨床上對骨刺的病情作判讀。並透過三維磁振MR影像重構出的膝關節模型做驗證，顯示出本系統在二維X光影像中偵測出的骨刺輪廓確實與利用三維MR影像重構出的膝關節模型的骨刺相符合。
在二維X光影像中，將影像分成股骨、膝關節、脛骨三個部份個別處理。在關節面的部分採用先定義出正常膝關節應該有的形狀，再向外偵測骨刺的策略。首先利用動態輪廓模型(Active Shape Model)，定義出初始膝關節輪廓，並利用以梯度(Gradient)為主的方法且配合輪廓邊緣附近區域(Region)、紋理(Texture)、外觀(Appearance)等資訊，以正常的膝關節輪廓為基礎搜尋出實際上膝關節真正的輪廓。最後利用此兩個輪廓的資訊整合病理學上的知識，套用模糊系統來偵測輪廓上骨刺發生的機率。而股骨與脛骨的部份，套用Canny 邊緣檢測將邊緣找出，並透過形態學擴張將不連續的邊連接起來，以推算出股骨軸與脛骨軸。最後計算出股骨軸與脛骨軸的斜率，以便與三維MR膝關節模型作對應。而在三維MR影像中，利用三個方向不同的切面，經過對位後，內插出整個體積資料，再挑選其中一個方向對整個體積資料進行切割。而後利用Marching Cube algorithm重構出膝關節模型，並以Inflation Method對網格模型的平滑性做適當的調整。在整合二維X光影像與三維MR膝關節模型方面，則利用MR影像重構出的膝關節模型，調整至與X光影像中膝關節相同角度、相同縮放比例時，即可觀察出在模型上的骨刺，確實與由系統在X光影像中偵測出的骨刺相吻合。

　The main purpose of this thesis is to establish a system, which detects bone spurs on the knee joint from two-dimensional X-ray images (AP view), to provide an objective basis for doctors to perform a powerful diagnosis. We also reconstruct the three-dimensional knee joint model from these stacks of two-dimensional MR volume images at three directions to test and verify whether the contours of bone spur we found conform to three-dimensional knee joint model we reconstruct.
　In the two-dimensional X-ray images, we divide the knee joint into three parts (femur, knee joint, and tibia) and process them individually. In the part of knee joint, we adopt the strategy to define the initial normal knee joint contour first, and then search the real knee joint contour outward. Next, we use the information of these contours and add the knowledge from pathology to detect bone spurs based on fuzzy logic control system. In the parts of femur and tibia, the border of femur and tibia is detected using canny edge operator and morphological dilation, and then we find the axis of femur and tibia. Finally, the slope of the axis for corresponding three-dimensional MR knee joint model is calculated. In the MR image, we use the slices from three directions to register and interpolate the volume data. We then pick up one direction to segment the volume data. At last, the knee joint model can be constructed by marching cube algorithm and refined by surface inflation.
　In order to compare the two-dimensional X-ray image with three-dimensional MR knee joint model, we have to adjust the angle and scale of the knee joint model to match the X-ray image. After alignment, we can observe that the bone spur on the 3D knee joint model and really conforms to the bone spur in the 2D X-ray images. We also design related experiments to calculate the corresponding error between three-dimensional knee joint model and two-dimensional X-ray images.

[1] C. B. Frank, R. M. Rangayyan, G. D. Bell, “Analysis of knee joint sound signals for non-invasive diagnosis of cartilage pathology” IEEE Engineering in Medicine and Biology Magazine, Volume 9, Issue 1, pp.65 – 68,1990

[2] S. A. Mazzuca, K. D. Brandt et. al., “Effect of alignment of the medical tibial plateau and X-Ray beam on apparent progression of osteoarthritis in the standing anteroposteior knee radiograph,” Arthritis & Rheumatism, vol. 44, no. 8, pp.1786-1794, 2001.

[3] M. J. Beynon, L .Jones, C. A. Holt, “Classification of osteoarthritic and normal knee function using three-dimensional motion analysis and the Dempster-Shafer theory of evidence,” IEEE Trans. on Systems, Man and Cybernetics Vol. 36, Issue 1, pp.173-186,2006

[4] A. Antani, L.R. Long, G.R. Thoma, R.J. Stanley, “Vertebra shape classification using MLP for content-based image retrieval,” Proceedings of IJCNN, vol.1, pp.160-165, July 2003.

[5] M. Cherukuri, R.J. Stanley, L.R. Long, A. Antani, G..R. Thoma, “Anterior osteophyte discrimination in lumbar vertebrae using size-invariant features,” Computerized Medical Imaging and Graphics, vol.28, pp99-108, 2004.

[6] Y. Chen, X. Ee, W. K. Leow and T. S. Howe, “ Automatic Extraction of Femur Contours from Hip X-ray Images,“ In Proc. First International Workshop on Computer Vision for Biomedical Image Applications(CVBIA 2005) ,pp. 200-209

[7] M. Kass, A. Witkin, and D. Terzopolous, ”Snake: Active Contour Models,” Int. J. Comput. Vis. 1, pp. 321-331, 1988.

[8] C. Xu, J.L. Prince,”Gradient vector flow: A new external force for snakes,” In Proc. IEEE Conf. on CVPR,1997

[9] Y. Wang and H. Lawrence, “Boundary Finding with Prior Shape and Smoothness Models,“ IEEE Trans. Patttern Analysis and Machine Intelligence, vol. 22, no. 7, July 2000.

[10] T.F. Cootes, D. Cooper, C.J. Taylor and J. Graham, “A Trainable Method of Parametric Shape Description” In Procs. British Machine Vision Conference, Springer, Verlag, pp.54-61, 1991.

[11] T.F. Cootes, C. J. Taylor., “Active shape models – smart snakes,” In Procs. British Machine Vision Conference, Springer, Berlin, pp. 266-275, 1992.

[12] T. F. Cootes, C. J. Taylor, D. H. Cooper, et al., “Active Shape Models - Their Training and Application,” Computer Vision and Image Understanding, pp.38-59, January 1995.

[13] T.F. Cootes, G.J. Edwards and C.J. Taylor, “Active appearance models,“ IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 23, no. 6, June 2001.

[14] F. Maes, P. Suetens and P. Dewaele, “Statistical shape model-based segmentation of digital X-ray images,“ In Procs. IEEE Workshop on Mathematical Methods in Biomedical Image Analysis 11-12, pp.61 – 68 ,June 2000

[15] M.Seise, S. J. McKenna, I.W. Ricketts and C.A.Wigderowitz , “Segmenting Tibia and Femur from Knee X-ray Images, “In Procs. Medical Image Understanding and Analysis (MIUA), Bristol, July 2005, pp.103-106.

[16] 洪啟峰，三維膝關節電腦模型之建立與應用，碩士論文，國立台灣大學醫學 工程學研究所，2003。

[17] Y. T. Chen and M. S. Wang, “Three-Dimensional Reconstruction and Fusion For Multi-modality Spinal Images,” Computerized Medical Imaging and Graphics, vol. 28, no. 1-2, pp. 21–31, Jan. 2004.

[18] X. Zhao, F. Qi, H. Huang, and J. Zhan, “3-D deformable models for femur shape recovery from volumetric images,” in Procs. The Fourth International Conference on High Performance Computing in the Asia-Pacific Region, vol. 2, pp. 797-798, May 2000

[19] A. B. Albu, D. Laurendeau, L. J. Hebert, H. Moffet, and C. Moisan, “Three-dimensional Reconstruction of the Bony Structures Involved in the Articular Complex of the Human Shoulder Using Shape-based Interpolation and Contour-based Extrapolation,” in Proc. Fourth International Conference on 3-D Digital Imaging and Modeling (3DIM), pp. 370-377, 2003.

[20] M. E. Tremblay, A. B. Albu, L. Hebert, D. Laurendeau, “Integrating Region and Edge Information For the Automatic Segmentation of Interventional Magnetic Resonance Images of the Shoulder Complex,” in Proc. First Canadian Conference on Computer and Robot Vision, pp. 279-286, May 2004.

[21] S.S. Rao, “Engineering Optimization: Theory and Practice,” Third edition, Wiley-Interscience, 1996.

[22] H. Embrechts, D. Roose, ”A Parallel Distance Transformation Algorithm,” Distributed Memory Computing Conference, 1991. Proceedings., The Sixth April 28-May 1, pp. 479 – 482, 1991.

[23] N. Otsu, “A threshold selection method from gray-level histogram,” IEEE Transaction on System, Man, and Cybernetics, Vol. 9, No. 1, pp. 62-66, January 1979.

[24] W.E. Lorensen, and H.E. Cline, “Marching Cubes: a high resolution 3D surface reconstruction algorithm,” Computer Graphics (Proc. of SIGGRAPH), Vol. 21, No. 4, pp. 163-169, 1987.

[25] H. C. Lee, Y. N. Sun, “Computer X-Ray Image Reconstruction for 3D Knee Joint and Computer-Assisted Diagnosis for Knee Osteoarthritis,” 國立成功大學資訊工程所碩士論文,2005

[26] Y. C. Chuang, Y. N. Sun, “Computerized 3D Object Reconstruction and Segmentation from Multi-spectral MR Images,” 國立成功大學資訊工程所碩士論文,2005

[27] R. C. Gonzalez, R. E. Woods, ”Digital Image Processing,” Second Edition, Prentice Hall

[28] 鄭慶明, “實用影像診斷學,” 俊傑書局發行, 2001.

[29] 楊英魁,孫宗瀛,鄭魁香,林建德,蔣旭堂, “模糊控制理論與技術,” 全華科技圖書股份有限公司, 2002.

[30] http://www.dartmouth.edu/~anatomy/knee/radiographs/radio2.html