下肢不等長(LLD)是兒童骨科門診與手術常見問題，而引發LLD常見的三種疾病分別為股骨頭缺血性壞死(LCPD)、股骨頭生長板脫位(SCFE)以及發展性髖臼發育不良(DDH)。LLD的臨床診斷最常使用的方式為X光檢查，醫生手動在下肢AP view 的X光影像上選取股骨與脛骨之特定結構位置點進行相關數據的量測，如股骨長度、脛骨長度、全腳長度以及股骨軸與脛骨軸之夾角。由於此判斷方式較為主觀，並且其準確度容易受到外在因素的影響(如精神疲勞)，進而導致錯誤的量測發生。在本論文研究中，我們將透過醫學影像處理技術對下肢AP view 的X光影像進行相關數據量測與分析。研究中所使用的影像充滿著許多雜訊，因此需先經過幾個濾波器進行前處理將雜訊移除並且強化邊緣資訊以利於後續的影像分割與偵測。在骨軸偵測方面，利用影像邊界資訊偵測股骨軸與脛骨軸，而在骨關節面分割方面則是利用動態形狀模型Active Shape Models分割股骨上下緣、脛骨上緣以及距骨關節面輪廓，同時利用了距骨的資訊進行脛骨下關節面的偵測。本研究並提出了一個微調輪廓的方法，提升分割結果的準確率。最後利用骨軸與骨關節面資訊進行相關數據的量測與評估，藉此輔助醫師更加了解病人的病況，進而給予適合的治療方式。

Leg length discrepancy (LLD) is a common disease in clinical orthopedics surgery and pediatric orthopaedic clinic. Legg-Calve-Perthes disease(LCPD), Slipped Capital Femoral Epiphysis(SCFE) and Developmental Dysplasia of the Hip(DDH) may induce LLD. Generally, orthopedic surgeons adopt X-ray image in LLD examination. From the acquired X-ray image, orthopedic surgeons select feature points manually in the lower limbs to measure some parameters, such as tibial length, femoral length, the length of the whole limb and the angle between femoral axis and tibial axis. Since the position judgment of those feature points highly depends on the subjective experience of surgeon, the detected feature points may be unstable. In this thesis, a robust system has been developed to measure the parameters of the lower limbs in X-ray image by using image processing technology. Since the X-ray image is always noisy, we must use some filters for noise reduction and edge enhancement as the preprocessing step to the images that will later be used in the image segmentation and measurement. In image segmentation, we first use the edge information of femur and tibia for finding the bone axis, and then adopt the Active Shape Model to segment the proximal femur surface, the distal femur surface, the tibial plateau surface and the talar dome surface. Similarly, the tibial plafond surface will be detected by segmenting the talar dome surface with ASM. We also propose a surface refinement step to enhance the accuracy of surface segmentation. Related clinical parameters mentioned above will be measured by our system automatically. These parameters could assist orthopedic surgeons to better understanding the illness of patients.

[1]Inan, M., G. Chan, and J.R. Bowen, “The correction of leg length discrepancy after treatment in developmental dysplasia of hip by using a percutaneous epiphysiodesis,” Journal of pediatric orthopedics, Part B, vol. 17, pp. 43-46, 2008.
[2]Kalamchi, A. and G.D. MacEwen, “Avascular necrosis following treatment of congenital dislocation of the hip,” The Journal of bone and joint surgery, vol. 62, pp.876-888, 1980.
[3]Schoenecker, P.L and W.B. Strecker, ”Congenital dislocation of the hip in children. Comparison of the effects of femoral shortening and of skeletal traction in treatment,” The Journal of bone and joint surgery, vol. 66, pp. 21-27, 1984.
[4]Grzegorzewski, A., et al., “Leg length discrepancy in Legg-Calve-Perthes disease,” Journal of pediatric orthopedics, vol. 25, pp. 206-209, 2005.
[5]Sanjeev Sabharwal MD, Ajay Kumar MD, “Methods for Assessing Leg Length Discrepancy,” Clinical Orthopaedics and Related Research, vol. 466, pp. 2910-2922, 2008.
[6]Sabharwal S, Zhao C, McKeon JJ, McClemens E, Edgar M, Behrens F, “Computed radiographic measurement of limb-length discrepancy. Full-length standing anteroposterior radiograph compared with scanogram,” The Journal of bone and joint surgery, vol. 88, pp. 2243-2251, 2006.
[7]M. Kass, A. Witkin, and D. Terzopoulos, “Snakes: Active contour models,” International journal of computer vision, vol. 1, pp. 321-331, 1988.
[8]C. Xu and J.L. Prince, “Gradient Vector Flow: A New External Force for Snakes,” Proc. IEEE Conf. on Comp. Vis. Patt. Recog. (CVPR), June 1997.
[9]T. F. Cootes, D. H. 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.
[10]T. F. Cootes and C. J. Taylor, “Active shape models – smart snakes,” In Procs. British Machine Vision Conference, Springer, Berlin, pp. 266-275, 1992.
[11]T. F. Cootes, C. J. Taylor, D. H. Cooper, et al, “Active Shape Models – Their Training and Application,” Computer Vision and Image Understanding, vol. 61, pp. 38-59, January 1995.
[12]T. F. Cootes, G. J. Edwards and C. J. Taylor, “Active appearance models,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 23, no. 6, pp. 681-685, June 2001.
[13]W. Xie, J. Franke, C. Chen, et al, “Statistical model-based segmentation of the proximal femur in digital antero-posterior (AP) pelvic radiographs,” International journal of computer assisted radiology and surgery, vol. 9, pp. 165-176, 2014.
[14]C. C. Chen and Y. N. Sun, “Automatic Image Analysis for Detection and Evaluation of Pelvic Diseases,” 國立成功大學資訊工程所碩士論文, 2007.
[15]M. Seise, S. J. McKenna, I. W. Ricketts and C. A. Wigderowitz, “Learning Active Shape Models for Bifurcating Contours,” IEEE Trans. On Medical Image, vol. 26, pp. 666-677, May 2007.
[16]H. C. Lee and 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.
[17]C. C. Huang and Y. N. Sun, “Bone Spur Detection and Evaluation for Knee Joint from 2D X-Ray and 3D MR Images,” 國立成功大學資訊工程所碩士論文, 2006.
[18]Hugo Embrechts and Dirk Roose, “A Parallel Distance Transformation Algorithm,” Distributed Memory Computing Conference, 1991.
[19]Nobuyuki Otsu, “A threshold selection method from gray-level histograms,” IEEE Trans. Sys., pp. 62-66, 1979.
[20]https://en.wikipedia.org/wiki/Histogram_equalization
[21]W. H. Tsai and Y. N. Sun, “Model-based Orthodontic Assessments from Dental Panoramic Radiographs,” 國立成功大學資訊工程所碩士論文, 2014.
[22]C. Tomasi and R. Manduchi, “Bilateral Filtering for Gray and Color Images,” IEEE International Conference on Computer Vision, Bombay, India, 1998.
[23]http://radiologymasterclass.co.uk/tutorials/musculoskeletal/x-ray_trauma_lower_limb/ankle_fracture_x-ray.htm