臺灣博碩士論文加值系統

摘 要
目前子宮抹片影像的診斷主要仰賴病理醫師以人工方式判讀，人工判讀易因精神倦怠及病理醫師經驗不足而產生錯誤，利用電腦輔助判讀不但可減少人為疏忽，更能提高判讀的效率。抹片判讀的依據為細胞核是否變大、形狀不規則、核質比是否過大及核染顏色較深等。利用電腦判讀抹片時，除核染顏色外，其餘判讀依據皆有賴準確的邊界偵測技術，準確的邊界偵測技術將可提高電腦判讀抹片的可行性。
傳統的影像邊界偵測方法在背景及物體的影像皆很均勻時，可以得到很好的效果。但當影像雜訊很大時，將面臨繁瑣及不可靠的邊界點連結。主動輪廓模型(Active Contour Model)已被証明可有效的改善傳統方法的缺點且已廣泛地被應用於不同領域之邊緣偵測及物體分割。傳統的主動輪廓模型其初始輪廓必須很接近物體輪廓以避免收斂至錯誤的區域最小能量值；而且在邊界輪廓內凹處並無法收斂至適當的邊界。梯度向量流主動輪廓有效的改善了傳統蛇型輪廓的缺點，大部份影像只要適當地調整其相關之各項參數值，皆可得到良好的偵測結果。本論文利用梯度向量流主動輪廓模型偵測抹片細胞之細胞核和細胞質輪廓，並探討梯度向量流模型中各項參數最佳的偵測值。研究結果顯示，偵測細胞質邊界之形變參數設成α=0.05、β=2、γ=5、κ=2，偵測細胞核之形變參數設成α=0.05、β=6、γ=4、κ=2，則針對不同影像只要調整σ、μ參數，皆可得到良好的偵測結果。最後以上述各參數設定偵測正常細胞及異常細胞之細胞質和細胞核邊界，並計算其核質面積比，結果顯示各細胞核質比值與理論值相符，驗証了偵測結果的正確性，梯度向量流主動輪廓技術的確可準確偵測抹片細胞輪廓。

Abstract
Accuracy of Pap smear diagnosis highly depends on the pathologists and their experiences. Using computers as an assistance is expected to increase the diagnostic accuracy. The need of an efficient boundary detection algorithm plays an important role in developing an useful computer-assisted diagnostic system.
Traditional edge detection techniques are more useful for images with homogeneous foreground and background. If the images are corrupted with noise or artifacts, they are prone to generate segmented edges. Kass et al. in 1987 proposed a novel algorithm, called active contour model (ACM), which has been shown to be able to improve the detection of the contours. It has two shortcomings: (1) the initial contour must be placed close to the object. Otherwise it might be attracted to the contour with local minimum energy; (2) it is difficult to converge to concave boundary. Xu and Prince recently proposed an algorithm combining gradient vector flow (GVF) with Balloon ACM. Their method has been demonstrated to effectively solve the above problems by tuning model parameters for images with different characteristics.
In this thesis, GVF was applied for detecting the boundaries of cell membrane and nucleus for images obtained from Pap smear tests. Optimal model parameters were obtained and used for accurately detecting the cell boundaries. The results show that only two parameters, σandμ, need to be adjusted for different images, while other parameters can be fixed for membrane (α=0.05, β=2, γ=5, κ=2) and nucleus (α=0.05, β=6, γ=4, κ=2) . The nucleus/membrane area ratios for both normal and diseased cells calculated using our method were shown to be the same as the published data, we conclude that the GVF model can be used for clinical application in detecting cell boundary. Future works will develop an fully automatic system for easily and quickly discriminating the abnormal cells from the normal one.

目 錄
第一章 簡介 1
1.1 研究動機及目的 1
1.2 章節介紹 3
第二章 主動輪廓模型 5
2.1數學背景及影像處理基礎概念 5
2.1.1 極值問題(Extrema Problem) 5
2.1.2 變異微積分(Calculus of Variation) 6
2.1.3 光流(Optical Flow) 8
2.1.4 曲度之計算(Evaluation of Curvature) 9
2.2 蛇型輪廓模型(Snakes) 10
2.3 氣球輪廓模型(Balloons) 15
2.4 梯度向量流模型(Gradient Vector Flow) 16
第三章 材料和方法 19
3.1 研究影像說明 19
3.2模型參數 21
3.2.1 a參數 21
3.2.2 b參數 24
3.2.3 s參數 24
3.2.4 m參數 27
3.2.5 g與κ參數 27
3.3 梯度向量流主動輪廓偵測模型演算法 30
3.4 研究方法 32
3.4.1 計算細胞影像的影像力Eext(edge map) 32
3.4.2 根據細胞影像力，計算光流向量V 40
3.4.3初始輪廓位置設定及形變參數的調整 40
第四章 結果與討論 53
第五章 結論 68
參考文獻 72

參考文獻
1. R. F. Gonzalez and P. Wintz, “Digital Image Processing, 3rd Ed.” Addison-Wesley 1992.
2. B. K. P. Horn, “Robot Vision,” MIT Press 1986.
3. C.-M. Chen, H. Lu, and Y.-C. Lin, “An early vision-based snake model for ultrasound image segmentation,” Ultrasound Med. Biol. Vol. 26, pp273-285, 2000.
4. G. Kovalski, R. Beyar, R. Shofti, and H. Azhari, “Three-dimensional automatic quantitative analysis of intravascular ultrasound images,” Ultrasound Med. Biol. Vol. 26, pp527-537, 2000.
5. F. Leymarie and M. D. Levine, “Tracking deformable objects in the plane using an active Contour model,” IEEE Trans. Patt. Anal. Machine Intell., vol 15, pp617-634, 1993.
6. C. A. Davatzikos and J. L. Prince, “An active contour model for mapping the cortex,” IEEE Trans Med. Image, vol 14, pp65-80, 1995.
7. C. Xu and J. L. Prince, “Snakes, shapes, and gradient vector flow,” IEEE Trans Image Proc., vol 7, pp359-369, 1998.
8. D. J. Williams and M. Shah, “A fast algorithm for active contours and curvature estimation,” CVGIP:Image Underst., vol 55, pp14-26, 1992.
9. C. Xu, D. L. Pham, M. E. Rettmann, D. N. Yu, and J. L. Prince, “Reconstruction of human cerebral cortex from magnetic resonance images,” IEEE Trans Med. Image, vol 18, pp467-480, 1999.
10. C. Xu, D. L. Pham, and J. L. Prince, “Image segmentation using deformable models,” Handbook of Medical Imaging: vol. 2. Medical Imaging Processing and Analysis, SPIE Press, pp129-174, 2000.
11. V. Caselles, R. Kimmel, and G. Sapiro, “Geodesic active contours,” Int. J. Comp. Vision, vol. 22, pp61-79, 1997.
12. K. Siddiqi, Y. B. Lauziere, A. Tannenbaum, and S. W. Zucker, “Area and length minimizing flows for shape sehmentation,” IEEE Trans Image Proc., vol 7, pp433-443, 1998.
13. C. Xu and J. L. Prince, “Gradient Vector Flow:A New External force for Snakes” IEEE Proc. Conf. on Comp. Vis. Patt. Recog.(cvpr’97).
14. 長庚紀念醫院曾志仁醫師，”子宮頸癌與子宮頸癌前期的防治” 中華民國衛生署子宮頸癌防治教材 http://www.cgmh.org.tw/intr/intr5/c6700/Dr2/Work/Teaching/Patient/子宮頸抹片衛教.htm
15. 鐘坤井醫師，”細胞病理學” http://www.cgmh.com.tw/intr/intr5/c6700/teaching/professional/婦科細胞病理學.htm