臺灣博碩士論文加值系統

磁振造影(Magnetic Resonance Imaging, MRI)有助於骨質疏鬆性脊椎壓迫性骨折(Osteoporotic Vertebral Compression Fractures)嚴重程度之診斷，通常從矢狀面較能清楚察看脊椎縱切面結構及影像顯影的變化。脊椎壓迫性骨折的初期症狀，常被誤認為僅是一般腰背疼痛或是舊傷的復發而被忽略；因未能及時被發現，導致持續惡化，甚至駝背畸形，嚴重者導致癱瘓。近年來骨質疏鬆性脊椎壓迫性骨折發生率逐漸攀升，因而增重了骨科醫師的工作量。當骨科醫師一天診查數百張MRI影像時，容易產生視覺疲勞而忽略疾病細節，因而造成誤診情況，也可能因此產生醫療糾紛。
為減輕醫師的工作量，進而減少誤診率，本篇論文目的在發展一以矢狀面MRI為基礎新壓迫性骨折影像辨識系統(MRI Based New Sagittal Vertebral Fracture Detection System, MRI-NSVF Detection System)，此系統能準確切割出以矢狀面T1加權模式MRI影像中每一節脊椎椎體(Vertebral Body)，並自動辨識此椎體是否為新壓迫性骨折；其可輔助醫師對新壓迫性骨折進行判讀。實驗結果顯示此系統對於新壓迫性骨折，能達91.58%正確診斷率。此外本論文亦提出一基因演算法，以推導出MRI-NSVF Detection System所採用之全部參數的最適值。

The severity grade of osteoporotic vertebral compression fractures (VCFs) can be examined by Magnetic Resonance Imaging (MRI) according to variable signals change of spine on sagittal plane. At the early stage of the osteoporotic VCFs, it is usually mistaken for general back pain or old trauma, leading to neglect fracture and continuing to be worsen, or even kyphotic deformity and paralysis in severe cases. In recent years, the cases of osteoporotic VCFs have gradually risen; increasing the workload of orthopedists and medical disputes from diagnostic errors resulted from orthopedists prone to visual fatigue and ignore the details of the disease after checking hundreds of MRI images in a day.
Therefore, in order to reduce misdiagnosis rate by effectively reducing the workload of orthopedists; an MRI Based New Sagittal Vertebral Fracture Detection System (MRI-NSVF Detection System) is developed in this thesis, based on the T1-weighted sagittal plane of MRI to accurately segment every level of vertebral body and automatically detect whether the vertebral body suffers from new compression fractures or not. The MRI-NSVF Detection System can assist orthopedists in diagnosing new sagittal vertebral fracture. The experimental results show that the MRI-NSVF Detection System can provide 91.58% of accuracy rate in diagnosing new VCFs detection. Moreover, in this thesis, a genetic algorithm is also proposed to derive the most appropriate values of all the parameters used in the MRI-NSVF Detection System.

摘要 i
Abstract ii
目錄 iii
表目錄 v
第一章 緒論 1
1.1. 研究背景與動機 1
1.2. 研究目的 3
1.3. 論文架構 3
第二章 文獻探討 5
2.1 壓迫性骨折相關研究 5
2.1.1 半定量法評估脊椎骨折(Vertebral Fracture Assessment Using a Semiquantitative Technique) 5
2.1.2 陳舊性癒合之壓迫性骨折、螺釘植入物及骨水泥之干擾現象 6
2.2斷開與閉合(Opening and Closing) 10
2.3 MRI脊椎影像自動化椎體偵測及切割(Automated Vertebra Detection and Segmentation from the Whole Spine MR images) 11
2.4使用紋理及灰度特徵半自動化分類於MRI影像之良性與惡性椎體壓迫性骨折(Semiautomatic classification of benign versus malignant vertebral compression fractures using texture and gray-level features in magnetic resonance images) 13
第三章 椎體切割方法 14
3.1 預處理階段 14
3.1.1 移除雜訊及過亮、過暗之區域 14
3.1.2 多數濾波器(Majority Filter) 17
3.1.3增強影像對比度 18
3.2椎體切割階段 19
3.2.1基於共變異數之梯度計算法(Covariance Based Gradient Computing Method ,CBGC) 20
3.2.2增強方法－Run-length enhancer及Otsu’s method 22
3.2.3 HMTS演算法(Hit-and-Miss Transform-based Skeletonization) 25
3.2.4移除錯誤區域 27
第四章 椎體切割改善方法 29
4.1預處理 29
4.1.1改良式直方圖均衡化(Modified Histogram Equalization) 30
4.1.2 增強方法 31
4.2椎間盤與組織的切割階段 32
4.2.1區域門檻值 32
4.2.1.1 門檻值步驟 33
4.2.1.2 合併步驟 35
4.2.2 改良增強方法 36
4.2.3 雙重區域門檻值 36
4.2.4 斷開與閉合(Opening and Closing) 38
4.2.5 移除殘留組織區域 39
4.2.6 檢查模式 42
第五章 辨識新壓迫性骨折方法 43
5.1 連通標記法 (Connected Components Labeling) 43
5.2 產生遮罩 45
5.2.1已切割出椎體之區域 45
5.2.2未切割出椎體之區域 47
5.3 GLCM紋理特徵計算 50
5.3.1灰度共生矩陣(Gray Level Co-occurrence Matrix，GLCM) 50
5.3.2 GLCM量化特徵值 51
5.4 計算距離分類 53
第六章 基於基因演算法之參數設定 55
6.1基於基因演算法之參數設定(GBPD) 55
6.2 椎體切割及辨識之參數設定 56
6.2.1 椎體切割方法之參數設定 56
6.2.2 椎體切割改善方法之參數設定 57
6.2.3 辨識新壓迫性骨折方法之參數設定 58
第七章 實驗結果 60
7.1 椎體切割結果與正確率 60
7.2 辨識新壓迫性骨折結果與正確率 64
第八章 結論與未來展望 67
參考文獻 68

[1] 張志儒醫師, 壓迫性骨折, 國泰綜合醫院神經外科醫療團隊. Retrieved April 11, 2016, from: http://www.cgh.org.tw/tw/content/depart/C300/DL/B0000000119.pdf
[2] 2015台灣成人骨質疏鬆症防治之共識及指引, 中華民國骨質疏鬆症學會. Retrieved April 22, 2016, from: http://www.toa1997.org.tw/download/2015Taiwan consensus guidelines and prevention of osteoporosis in adults.pdf
[3] 鄧木火醫師, 骨質疏鬆椎體骨折的治療簡介, 台北榮民總醫院放射線部神經放射科. Retrieved April 25, 2016, from: http://homepage.vghtpe.gov.tw/~rad/VP/VPnew.htm
[4] 骨質疏鬆與骨折之治療, 中國醫藥大學附設醫院骨科部. Retrieved March 2, 2017, from: http://www.cmuh.cmu.edu.tw/web/showhealthinfo.php?id=383
[5] T.A. Predey, L.E. Sewall, S.J. Smith. "Percutaneous Vertebroplasty: New Treatment for Vertebral Compression Fractures", Am Fam Physician. Vol. 66, No. 4, pp. 611-616, 2002.
[6] 陳志宏教授, 張允中醫師, 謝昭賢博士, 核磁共振與磁振造影, 台灣大學電機系核磁共振影像光譜實驗室,台灣大學醫學院附設醫院影像醫學部,台灣大學電機系核磁共振影像光譜實驗室. Retrieved April 15, 2016, from:
http://www.csie.ntu.edu.tw/~b95041/biomedical/11_16_MR_Simple.pdf
[7] 高崢嶸, 徐新峰, "MRI脂肪抑制(STIR)序列在外傷性腦脊液漏中的應用", Chinese Community Doctors, Vol. 11, No. 22, 2009.
[8] 莊雅方, "以MRI影像為基礎之脊椎節自動切割方法", 碩士論文, 國立中興大學資訊管理學系所, 2012。
[9] 洪愷均, "建立在 MRI 影像之基礎的脊椎自動切割方法", 碩士論文, 國立中興大學資訊管理學系所, 2014。
[10] R.M. Haralick, K. Shanmugam, and I. Dinstein. "Textural features for image classitication", IEEE Transactions on Systems, Man, and Cybernetics, Vol. 3, No. 6, pp.610-621, 1973.
[11] Z. Peng, J. Zhong, W. Wee, et al. "Automated Vertebra Detection and Segmentation from the Whole Spine MR Images", Proceedings of IEEE's Engineering in Medicine and Biology Society, Vol. 3, pp. 2527-2530, 2005.
[12] L. Frighetto-Pereira, R. Menezes-Reis, G. A. Metzner, et al. "Semiautomatic classification of benign versus malignant vertebral compression fractures using texture and gray-level features in magnetic resonance images", IEEE 28th International Symposium on Computer-Based Medical Systems, pp. 88-92, 2015.
[13] H.K. Genant, C.Y. Wu, V. Kuijk C, et al. "Vertebral fracture assessment using a semiquantitative technique", Journal of Bone and Mineral Research, Vol. 8, pp. 1137-1148, 1993.
[14] J.M. Bassano. "Lumbar Spine: Old vs. New Compression Fracture", Dynamic Chiropractic, Vol. 24, 2006.
[15] D.B. Scott. "Diagnosing Vertebral Compression Fractures". Retrieved March 10, 2017, from: http://www.spine-health.com/conditions/osteoporosis/diagnosing-vertebral-compression-fractures
[16] J. Canny. "A Computational Approach To Edge Detection", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 8, No. 6, pp. 679-698, 1986.
[17] S.F. Yang-Mao, Y.K Chan, Y.P. Chu. "Edge Enhancement Nucleus and Cytoplast Contour Detector of Cervical Smear Images", IEEE Transactions on Systems, Man and Cybernetics-Part B, Vol. 38, No. 2, pp. 353-366, 2008.
[18] Y.K Chan, D.C. Huang, K.C. Liu, et al. "An Automatic Indirect Immunofluorescence Cell Segmentation System", Hindawi Publishing Corporation Mathematical Problems in Engineering, Vol. 2014, Article ID 501206, 2014.
[19] L. Lam, S.W. Lee, and C.Y. Suen. "Thinning Methodologies-A Comprehensive Survey", IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 14, No. 9, pp. 869-885, 1992.
[20] R. Haralick and L. Shapiro. "Computer and Robot Vision", Addison-Wesley Publishing Company, Vol. 1, Chap 5, pp. 168-173, 1992.
[21] N. Otsu. "A Threshold Selection Method from Gray-Level Histograms", IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, No. 1, pp. 62-66, 1979.
[22] J.H. Holland. "Genetic Algorithms", Scientific American, Vol. 267, pp. 66-72, 1992.
[23] K. Alsabti, S. Ranka and V. Sigh. "An Efficient K-means Clustering algorithm", in Proc. 1st Workshop on High Performance Data Mining, 1998.
[24] C. Silva, B. Ribeiro, and N. Lopes, "Improving Recall Values in Breast Cancer Diagnosis with Incremental Background Knowledge", in The 2010 International Joint Conference on Neural Networks, 2010.
[25] J. Davis and M. Goadrich, "The Relationship Between Precision-Recall and ROC Curves", in Proceedings of the 23rd International Conference on Machine Learning, 2006.