臺灣博碩士論文加值系統

超音波影像的多目標分割，可以幫助臨床醫師分割出可疑的目標物，可以幫助檢測人員節省手繪目標物的時間，也可用於新進人員的訓練之用，並且在發展電腦輔助診斷的系統上，扮演著非常重要的角色。

但是，由於超音波影像並不容易識別，因為具有低對比、高雜訊、斑點、假影、穿透現象等等的問題。這些現象使得超音波影像的目標物邊界容易模糊而不易識別。為了能夠降低超音波本身物理特性的影響，而能夠找出腫瘤的邊界輪廓，並且能夠減少超音波腫瘤影像容易過度分割的情形，在本研究裡，提出了等時性演化為基礎的區域單元競爭演算法。

本研究提出的等時性演化之區域單元競爭演算法，可分為兩個主要步驟。首先利用影像濾波器，降低超音波影像的雜訊，並強化邊界的特徵。接著使用分水嶺轉換演算法，以過度分割的方式，找出所有可能的邊界，並產生初始的區域單元。在結束這樣的前置處理與分割之後，依據等時性演化規則，引導區域的成長與區域彼此間競爭的過程，而分割出影像之中的目標物。

每一個區域有其成長的速度，其速度由局部相似值與全域變異值所組成。局部相似值代表區域局部相似的程度，促使區域向外成長；全域變異值代表區域全域的相似程度，代表區域向外成長的阻力。

本研究在臨床超音波影像的實驗結果中，不僅能夠有效的分割出腫瘤目標物，並且利用時間微調的彈性，對於惡性腫瘤常存在的腫瘤目標物過度分割的情形，能夠有所改善，而且經由本研究的驗證方式，本研究方法分割腫瘤影像與專家手繪分割腫瘤影像的誤差小於不同專家之間的手繪誤差。

Boundary extraction of multiple targets in a sonogram can help clinicians find out perceptible objects, save time for sinologists and be used for training. Moreover, it plays an essential role in developing computer-aided diagnosis systems.
Because of the intrinsic properties in ultrasonic images, such as low contrast, high noises, speckle, artifacts and so on, it is generally difficult to automatically identify the boundaries of the images. These intrinsic properties make the desired edges blurred and deteriorate the discriminability of the boundaries. In order to alleviate the influence of these physical problems and mitigate the over-segmentation of targets, we proposed a new segmentation algorithm, which was a cell competition algorithm based on isochronal evolution.
The proposed method composed of two steps. First, we used the image processing filters such as Gaussian filter and Sobel filter to smooth the images and enhance the boundary information. After that, we used the Watershed transformation to over-segment the images to capture all possible boundaries and generate the initial cells. After completion of pre-segmentation, cell competition was performed following the criteria of isochronal evolution, which were imposed on regions when they were growing and competing with one another.
Each region grew with its own viscosity which consisted of local similarity and global variation. Local similarity serves as the primary force for region expansion, whereas global variation plays the role of expansion hindrance.
The experimental results on the clinical ultrasound images showed that our algorithm could identify all objects of interest reasonably well. Moreover by adjusting the beginning time, we can alleviate the over-segmentation for malignant lesions. The algorithm was validated by comparing with manually outlined boundaries. The results showed that the differences between the boundaries derived by the proposed method and the hand-outlined boundaries were within the range of the differences among different observers.

第一章 緒論 1
1.1背景與研究動機 1
1.2研究目的 2
1.3資料取得 2
1.4研究架構 2
第二章 文獻探討 3
2.1臨界值法（Threshold） 4
2.1.1整體臨界值法 4
2.1.2適應性臨界值法 4
2.2群聚分析（Clustering） 5
2.2.1K-平均法 5
2.2.2階層群聚法 5
2.3區域成長法（Region growing） 6
2.4區域單元競爭演算法（Cell competition algorithm） 7
2.4.1區域單元競爭規則 7
2.4.2合併與分割的限制條件 7
2.5形變模型（Deformable model） 8
2.5.1參數式形變模型 8
2.5.2幾何式形變模型 9
2.6等位函數(Level Set) 10
2.6.1傳統等位函數法 10
2.6.2穩定性等位函數法（stationary level set method） 12
2.6.3等位函數疊代方程式 14
2.6.4快速逕行法 16
2.6.5等位函數在影像分割上的應用 17
第三章：前置處理與分割 19
3.1影像高斯濾波器 19
3.2邊緣偵測濾波器 21
3.3統計矯正 22
3.4分水嶺分割演算法 23
3.4.1第一次分水嶺轉換 23
3.4.2第二次分水嶺轉 24
第四章：基於等時性演化之區域單元競爭演算法 26
4.1時間定義 26
4.2速度函數定義 27
4.2.1統計檢定 27
4.2.2局部相似值 28
4.2.3全域變異值 30
4.2.4正規化（Normalization） 32
4.4等時性演化之區域單元競爭規則（isochronal evolution rule ） 33
4.4.1單一區域演化規則(single region evolution rule) 33
4.4.2多區域等時性演化規則（multi-region isochronal evolution rule） 34
4.4.3競爭之合併與分割規則（merge and split rule） 35
4.4.3初始時間微調(initial time set up) 38
4.4.4循環式更新程序(update procedure) 39
第五章 結果與討論 40
5.1影像取得與參數設定 40
5.2實驗與討論 41
5.2.1良性乳房腫瘤實驗 42
5.2.2惡性乳房腫瘤實驗 83
5.2.3時間微調的影響實驗 100
5.2.4演算法分割結果與手繪分割之比較 105
第六章 結論與未來研究方向 106
參考文獻 108

[1] Adalsteinsson D and Sethian JA, “The fast construction of extension velocities in level set methods,” Journal of computational physics, vol. 148, pp. 2-22, 1999.
[2] Adams R and Bischof L , “Seed region growing,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 16, pp. 641-647, 1994.
[3] Bottigli U and Golosio B, “Feature extraction from mammographic images using fast marching methods,” Nuclear instruments and methods in physics research, vol. 487, pp.209-215, 2002.
[4] Chopp DL, “Computing minimal surface via level set curvature flow,” Journal of computational physics, vol.106, pp.77-91, 1993.
[5] Chalana V , Linker DT, Haynor DR and Kim Y , “A multiple active contour model for cardiac boundary detection on echocardiographic sequences,” IEEE Transactions on Medical Imaging, vol. 15, pp. 290-298, 1996.
[6] Gauch JM , “Image segmentation and analysis via multiscale gradient watershed hierarchies,” IEEE Transactions on Image Processing, vol. 8, pp. 69-79, 1999.
[7] Jacob M, Blu T and Unser M, “Efficient energies and algorithms for parametric snakes,” IEEE Transactions on image processing, vol.13, pp. 1231-1244, 2004.
[8] Kass M, Witkin AP and Terzopoulos D, “Snakes: active contour models,” International journal of computer vision, vol. 1, pp. 321-331. 1987.
[9] Malladi R, Sethian JA and Vemuri BC, “Shape modeling with front propagation: a level set approach,” IEEE Transaction on pattern analysis and machine intelligence, vol. 17, pp. 158-175, 1995.
[10] Osher S and Fedkiw RP, “Level set methods: an overview and some recent results,” Journal of Computational Physics, vol. 169, pp. 463-502, 2001.
[11] Osher S and Sethian JA , “Fronts propagating with curvature dependent speed: Algorithms based on Hamilton-Jacobi formulations,” Journal of computational physics, vol. 79, pp. 12-49, 1988.

[12] Sethian JA, “Level set methods and fast marching methods,” Cambridge Press, Cambridge, 1999.
[13] Sifakis E, Garcia C and Tziritas G, ” Bayesian level sets for image segmentation,” Journal of Visual Communication and Image Representation, vol. 13, pp. 44-64, 2002
[14] Sethian JA, “A fast marching level set method for monotonically advancing fronts,” Proceeding of the National Academy of Science, vol 93, pp. 1591-1595,1996.
[15] Vincent L and Soille P , “Watershed in digital spaces: An efficient algorithm based on immersion simulations,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 13, pp. 583-598, 1991.
[16] Wang D, “A multiscale gradient algorithm for image segmentation using watersheds,” Pattern recognition, vol. 30, pp.2043-2052, 1997.
[17] Xiaquan S and Spann M , “Segmentation of 2D and 3D images through a hierarchical clustering based on region modeling,” International Conference on Image Processing (ICIP ''97), vol. 3, pp. 50, 1997.
[18] Yan J, Zhuang T, Zhao B, and Schwartz LH,”Lymph node segmentation from CT images using fast marching mthods,” Computerized medical image and graphics, vol. 28, pp.33-38, 2004.
[19] Yan JY and Zhuang TG, “Applying improved fast marching method to endocardial boundary detection in echocardiographic images,” Pattern recognition letters, vol. 24, pp. 2777-2784, 2003.
[20] Zhu F and Tian J, “Modified fast marching and level set method for medical image segmentation,” Journal of x-ray science and technology, vol.11, pp. 193-204, 2003.
[21] Zhu SC, “Region competition: Unifying snakes, region growing, and Bayes/MDL for multi-band image segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, pp. 884-900, 1996.
[22] 陳博量.“能量區域化與具中隔保留能力之區域單元競爭演算法,”台大醫工所碩士論文,2004.
[23] 蕭安廷.”區域單元競爭演算法：超音波影像多目標分割技術,”台大醫工所碩士論文,2003.