甲狀腺(thyroid)是人體最重要的內分泌腺之一，而甲狀腺疾病為國人常見的疾病之一。由於甲狀腺體積的大小是衡量疾病發生的重要指標，藉由量測甲狀腺的體積，對於醫生在診斷上將會有很大的幫助。
電腦斷層掃描(computer tomography, CT)是現在臨床上用於診斷甲狀腺疾病常見的輔助工具之一。雖然甲狀腺區域可由電腦斷層影像中手動標記出來，但由醫師手繪出影像中的甲狀腺輪廓卻相當耗費時間與精神。因此，本論文提出一個在電腦斷層影像上的自動化甲狀腺切割與體積估測系統。首先，系統自動地擷取出甲狀腺CT影像感興趣的區域(region of interest, ROI)，並濾除其它不必要的非甲狀腺腺體。之後結合區域成長法(region growing)與學習向量量化(Learning Vector Quantization, LVQ)類神經網路，利用連續輸入影像間的相似性進行網路的訓練以及測試，以實現甲狀腺區域切割，並更進一步的估測其體積。由實驗結果顯示，本研究在甲狀腺區域切割與體積估測上均有良好的效果。

Thyroid is one of the most important endocrine of human, and thyroid diseases are common diseases for our national. Because thyroid volume is significant indicator for diagnosing diseases, it is greatly helpful for physicians to diagnose by measuring thyroid volumes.
Computer tomography (CT) is a common Computer Aided Diagnosis (CAD) tool for clinical thyroid diseases diagnosis. Nevertheless, even if the thyroid regions can acquire with hand-marked from CT images. The procedure of outlining the thyroid regions by physicians consumes a good deal of time and mind. In this paper, we propose an automatic thyroid segmentation system in CT images. Firstly, the region of interest (ROI) of thyroid form CT is extracted automatically, and unnecessary non-thyroid glands are filtered out. Then, we combine region growing with Learning Vector Quantization (LVQ) neural network, and use the similarities of consecutive images to train and test the neural network. Finally, we estimate the thyroid volume by segmentation results. The experimental results show the effectiveness of thyroid segmentation and volume estimation.

摘要 ------------------------------------------------------------------------------------------ i
ABSTRACT -------------------------------------------------------------------------------- ii
目錄 ------------------------------------------------------------------------------------------ iii
圖例索引 ------------------------------------------------------------------------------------ iv
表格索引 ------------------------------------------------------------------------------------ vi
第一章 緒論 ----------------------------------------------------------------------- 1
1.1 研究動機 ----------------------------------------------------------------- 1
1.2 醫學影像分割簡介 ----------------------------------------------------- 1
1.3 相關研究 ----------------------------------------------------------------- 3
1.4 研究方法 ----------------------------------------------------------------- 5
1.5 章節大綱 ----------------------------------------------------------------- 6
第二章 前處理 -------------------------------------------------------------------- 7
2.1 擷取甲狀腺ROI --------------------------------------------------------- 7
2.1.1 尋找氣管 ------------------------------------------------------------- 8
2.1.2 定義ROI大小 ------------------------------------------------------ 9
2.2 去雜訊 -------------------------------------------------------------------- 10
2.3 濾除其它腺體 ----------------------------------------------------------- 11
2.3.1 k-means分群演算法 ----------------------------------------------- 11
2.3.2 區域填充 ------------------------------------------------------------- 13
2.3.3 圓形結構元素侵蝕 ------------------------------------------------- 14
2.3.4 遮罩式標籤法 ------------------------------------------------------- 15
2.3.5 解剖學特徵計算 ---------------------------------------------------- 17
2.3.6 分數排名 ------------------------------------------------------------- 20
第三章 甲狀腺切割與體積估測 ----------------------------------------------- 21
3.1 甲狀腺切割 -------------------------------------------------------------- 22
3.1.1 區域成長法 ---------------------------------------------------------- 22
3.1.2 亮度校調 ------------------------------------------------------------- 24
3.1.3 特徵擷取 ------------------------------------------------------------- 25
3.1.4 學習向量量化類神經網路 ---------------------------------------- 30
3.2 甲狀腺體積估測 -------------------------------------------------------- 36
第四章 實驗結果與討論 -------------------------------------------------------- 38
4.1 影像資料與實驗環境 -------------------------------------------------- 38
4.2 甲狀腺切割準確性評估 ----------------------------------------------- 38
4.3 甲狀腺切割方法與其他方法比較 ----------------------------------- 40
4.4 體積估測準確性評估 -------------------------------------------------- 43
4.5 前處理探討 -------------------------------------------------------------- 44
4.6 特徵選取探討 ----------------------------------------------------------- 46
4.7 亮度校調探討 ----------------------------------------------------------- 48
4.8 分割結果展示 ----------------------------------------------------------- 49
第五章 結論 ----------------------------------------------------------------------- 53
參考文獻 ------------------------------------------------------------------------------ 54
附錄 Using Learning Vector Quantization Neural Network for Thyroid Segmentation and Volume Estimation in CT images 57

[1]Massol J, Pazart L, Aho S, Strauch G, Leclere J, Durieux P, ” Management of the thyroid nodule: preliminary results of a practice survey of 685 general practitioners and specialists,” Ann Endocrinol, Vol. 54, pp. 220-225, 1993.
[2]Bruneton JN, Balu-Maestro C, Marcy PY, Melia P, Mourou MY, ” Very high frequency (13MHz) ultrasonographic examination of the normal neck: detection of normal lymph nodes and thyroid nodules,” Journal of Ultrasound Med, Vol. 13, pp. 87-90, 1994.
[3]WHO, “Indicators for assessing iodine deficiency disorders and their control through salt iodization,” World Health Organization, Geneva, Switzerland, 1994.
[4]Brown MC, Spencer R., “Thyroid gland estimated by use of ultrasound in addition to scintigraphy,” Acta Radiol Oncol Radiat Phys Biol, Vol. 17, pp. 337 -341, 1978.
[5]Wormanns D, Kohl G, Klotz E, et al., “Volumetric measurements of pulmonary nodules at multi-row detector CT: in vivo reproducibility,” European Radiol, Vol. 14, pp. 86-92, 2004.
[6]Van Hoe L, Haven F, Bellon E, et al., “Factors influencing the accuracy of volume measurements in spiral CT: a phantom study,” Journal of computer assisted tomography, Vol. 21 pp. 332-338, 1997.
[7]N. Archip and P.-J. Erard et al., “A Knowledge-Based Approach to Automatic Detection of the Spinal Cord in CT Images,” IEEE Transactions on Medical Imaging, Vol. 21, No. 12, pp. 1504-1516, 2002.
[8]M.S. Brown, M.G. Micheal, and G.G. Jonathan, “Automated measurement of single and total lung volume from CT,” Journal of Computer Assisted Tomography, Vol. 23, No. 4, pp. 632-640, 1998.
[9]R. Pohle and K. D. Toennies, “Segmentation of medical images using adaptive region growing,” Opt. Eng, Vol. 4322, pp. 1337-1346, 2001.
[10]Kass, M., A. Witkin, D. Terzopoulos, “Snakes: Active contour models,” Int. Journal of Computer Vision, Vol. 1, pp. 321-331, 1987.
[11]Cvancarova M, Albregtsen F, Brabrand K,Samset E, “Segmentation of ultrasound images of liver tumors applying snake algorithms and GVF,” International Congress Series, Vol. 1281, pp. 218-223, 2005.
[12]Min Wei, Yongjin Zhou, and Minqxi Wan, “A fast snake model based on non-linear diffusion for medical image segmentation,” Computerized Medical Imaging and Graphics, Vol. 28, No. 3, pp. 109-117, 2004.
[13]J. B. MacQueen, "Some Methods for classification and Analysis of Multivariate Observations", Proceedings of 5-th Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, University of California Press, Vol. 1, pp. 281-297, 1967.
[14]K. Laws, ”Textures Image Segmentation,” Ph.D. Dissertation, University of Southern California, 1980.
[15]Chung-Ming Wu, Yung-Chang Chen, “Statistical feature matrix for texture analysis,” Proc. of CVGIP, Vol. 54, pp. 407-419, 1992.
[16]T. Kohonen, “Learning Vector Quantization for Pattern Recognition,” Technical Report TKK-F-A601, Helsinki University of Technology, Finland, 1986.
[17]T. Kohonen, “Improved Versions of Learning Vector Quantization,” Proc. of the International Joint Conference on Neural Networks, San Diego, CA, Vol. 1, pp. 545-50, 1990.
[18]J.C. Bezdek, “Fuzzy Mathematics in Pattern Classification,” Ph. D. Dissertation, Cornell University, Ithaca, New York, 1973.
[19]Haralick. RM, Shanugam K, and Dinstein I, “Texture Feature for Image Classification,” IEEE Transactions on Systems, Man and Cybernetics, Vol. 3, No. 6, pp.610-621, 1973.
[20]張芳元，淺談甲狀腺疾病的病癥與治療，http://www.chgh.org.tw/%E6%96%B0%E8%81%9E/90-%E8%AA%8D%E8%AD%98%E5%A4%A7%E8%84%96%E5%AD%90.htm
[21]趙一平，2002，肺臟斷層掃描之電腦輔助分析，中原大學醫學工程研究所，碩士論文。
[22]許家群，2006，使用邊緣資訊之區域成長演算法與互動式編輯工具在醫學影像分割上的應用，國立台灣大學醫學工程研究所，碩士論文
[23]黃美玲，2006，類神經網路輔助醫療診斷分類模式之建構，國立勤益技術學院工業工程與管理學系，中華民國品質學會第42屆年會暨第12 屆全國品質管理研討會。
[24]陳淑嬌，2003，梯度向量流主動輪廓模組於抹片細胞邊界偵測之應用，國立中興大學電機工程研究所，碩士論文。
[25]洪欽銘，陳瑄易，2005，結合K-means演算法之類神經-模糊分類器於醫學診斷之應用，第十三屆模糊理論及其應用研討會。
[26]莊達峰，2005，使用模糊化學習向量量化神經網路於鼻咽部復發腫瘤之偵測，國立雲林科技大學資訊工程研究所，碩士論文。