對胸部X光影像進行分析前，肺臟區域切割是很重要的第一步驟。過去有許多研究採用監督式切割方法(Supervised segmentation method)，得到很好的成果，但是監督式方法需要花大量的時間和人力，來取得所需要的訓練資料，並不適合實現此種方法於台灣醫院。因此，我們提出無監督式切割方法建立於多重解析度碎形特徵向量(Multiresolution fractal feature vector)。多重解析度碎形特徵向量是先利用M-頻段小波進行多重解析度分析，再用碎形維度來描述其特徵向量，此特徵向量已經在肝臟超音波影像上有可信賴的結果，證明為適合描述具有重複材質的影像。基於此特徵向量，我們可以用簡單的非監督式分群法，把肺臟區域的像素和非肺臟區域的像素分開，但會得到不完整而且邊緣破碎的結果，因此我們加入其他的切割方法來完成非監督式切割。
我們利用圖形切割演算法(Graph cut segmentation)來得到一個邊緣完整的初始肺臟區域輪廓，除此之外，我們在圖形切割演算法裡結合了解剖學知識來減少鎖骨的干擾，讓初始輪廓更加接近最終結果。最後，我們使用動態輪廓線模型，來得到最終的肺臟區域輪廓，同時也結合些許解剖學知識，讓最終的結果能更貼近真實的肺臟區域輪廓。我們應用此非監督式切割方法於兩種影像，一是來自醫院的影像，另外是網路上的資料庫(JSRT database以及對應正確解SCR database)。從醫院影像的結果可以看出此方法的可行性，而在網路資料庫的表現成效相當不錯，可以和這兩年建立於動態輪廓線模型的非監督式切割方法相互比較。值得一提的是，近年來建立於動態輪廓線模型的方法，大多利用解剖學知識來取得初始輪廓，容易受到干擾並且不穩定。而我們方法基於多重解析度碎形特徵向量，能較有效取得的初始肺臟區域輪廓，已利於完成後續的動態輪廓線模型。

Lung field segmentation in chest radiograph is an essential and important step for automatically analyzing x-ray image. There are various supervised methods that perform well. But the acquisition of training data is time-consuming and requires great effort for the clinicians of the Hospital in Taiwan. Hence, we present an unsupervised method based on multiresolution fractal feature vector. The feature vector consists of M-band wavelet transform and fractal geometry representation. Such a multiresolution fractal feature vector has been proved trustworthy in the application of texture images and liver ultrasound images. With the robust feature vector, the difference between lung field region and non-lung field region can be distinguished by simple unsupervised clustering. But the clustering result from multiresilution feature vector is rough and ragged, so we apply other segmentation methods to complete the unsupervised segmentation.
We apply graph cut segmentation with anatomy information to obtain a regional initial lung field contour without the interference of clavicle edges. In the end, the final contour is obtained by anatomy-based active contour model. We do some refinements for active contour model in order to fit the boundary of lung field better, especially at the bottom of lung field contour. In the experimental results, we apply our method on real case images and JSRT database with ground truth (SCR database). The results from real case show the feasibility of our method. And the performance of images from JSRT database is comparative to other unsupervised segmentation methods which were proposed two years ago. Deserved to be mentioned, there are some methods based on active contour model. But the acquisition of initial contour in our method is more robust than other methods which are based on rule-based method.

[1] World Health Organization. WHO report, “Global tuberculosis control, epidemiology, strategy, financing,” 2009
[2] J.R. Muhm, W.E. Miller, R.S. Fontana, D.R. Sanderson, and M.A. Uhlenhopp, “Lung cancer detected during a screening program using four-month chest radiographs,” in Radiology 148, 609--615, 1983
[3] A. Schilham,, B. van Ginneken, and M. Loog , “A computer-aided diagnosis system for detection of lung nodules in chest radiographs with an evaluation on a public database,” in Medical Image Analysis 10 (2), 247--258.,2006
[4] B. van Ginneken, B. ter Haar Romeny, and M. A. Viergever, “Computer-aided diagnosis in chest radiography: A survey,” IEEE Transactions on Medical Imaging, vol. 20, no. 12, pp. 1228--1241, 2001.
[5] B. van Ginneken, M. B. Stegmann, and M. Loog, “Segmentation of anatomical structures in chest radiographs using supervised methods: comparative study on a public database,” in J. Med. Image Anal , 10, 19--40, 2006.
[6] W.-L. Lee and K.-S. Hsieh, “A robust algorithm for the fractal dimension of images and its applications to the classification of natural images and ultrasonic liver images,” in Signal Processing .9 (6), 1894--1904, 2010.
[7] Kiwon Sohn, “Segmentation of lung fields using Chan-Vese active contour model in chest radiographs,” in Proc. SPIE 7963, 796332 , 2011.
[8] P. Annangi, S. Thiruvenkadam, A. Raja, H. Xu, Xi Wen-Sun and Ling Mao, “A Region Based Active Contour Method For X-ray Lung Segmentation Using Prior Shape And Low Level Features,” in Biomedical Imaging: From Nano to Macro, 2010 IEEE International Symposium on, 892--895, 2010.
[9] B. van Ginneken and B. ter Haar Romeny, “Automatic segmentation of lung fields in chest radiographs,” in Medical Physics 27 (10), 2445--2455, 2000.
[10] T.F. Cootes and C.J. Taylor, “Statistical models of appearance for computer vision,” Technical Report, Wolfson Image Analysis Unit, University of Manchester, 2001.
[11] T.F. Chan and L.A. Vese, “Active contours without edges,” in IEEE Trans Image Proc 10 (2), 266--277, 2001.
[12] S. Mallat, “A theory for multiresolution signal decomposition: the wavelet representation,” in IEEE Trans. Pattern Anal. and Machine Intell., vol. 11, pp.674-693, July 1989.
[13] T. Chang and C. Kuo, “Texture analysis and classification with tree-structured wavelet transform,” in IEEE Trans. Image Processing, vol. 2, pp.429--441, Oct. 1993.
[14] Wen-Li Lee, Yung-Chang Chen, and Kai-Sheng Hsieh, “Ultrasonic Liver Tissues Classification by Fractal Feature Vector based on M-band Wavelet Transform,” in IEEE Trans. Med. Imag. 22(3), 382--392, 2003
[15] N. Sarkar and B. B. Chaudhuri, “An efficient differential box-counting approach to compute fractal dimension of image,” in IEEE Trans. Systems Man and Cybernetics, vol. 24, pp. 115--120, Jan. 1994.
[16] B. B. Mandelbrot, “Fractal Geometry of Nature,” in Freeman Press, San Francisco,1982.
[17] Y. Boykov and G. Funka-Lea, “Graph cuts and efficient n-d image segmentation,” in International Journal of Computer Vision, 70(2), 109--131, 2006
[18] M. Kass, A. Witkin, and D. Terzopoulos, “Snakes: Active contour models,” in Int. J. Computer Vision. 1(4), 321—331, 1987
[19] C. Xu and J. L. Prince, “Gradient vector flow : A new external force for snakes,” in IEEE Proc. Conf. on Computer Vision and Pattern Recognition, 66—71, 1997
[20] J. Shiraishi, S. Katsuragawa, J. Ikezoe, T. Matsumoto, T. Kobayashi, K. Komatsu, M. Matsui, H. Fujita, Y. Kodera, and K. Doi, “Development of a digital image database for chest radiographs with and without a lung nodule: receiver operating characteristic analysis of radiologists' detection of pulmonary nodules,” in American Journal of Roentgenology, 174, 71—74, 2000