臺灣博碩士論文加值系統

本論文的目標是發展一個有效的CAD演算法在肺腫瘤的偵測上，我們偵測的焦點為肺部血管樹中的腫瘤，包括附著於血管的腫瘤及懸浮於血管附近的腫瘤。為了達成這個目標，在本論文中，提出一個新的三維 (Three-dimensional, 3D) 肺腫瘤偵測演算法，是根據一個稱為形態動態 (Morphological Dynamics) 的新概念下而形成的。
為了幫助醫生達到肺腫瘤的高偵測率及低誤報率，過去中，不同種類的電腦輔助偵測 (Computer-aided Detection, CAD) 演算法被開發，包括一些商用的CAD系統及眾多的實驗室演算法。但是，在這些CAD演算法中，沒有任何一個演算法達到滿意的效能，它們共同的缺陷為靈敏度的不足、高誤報率以及無法偵測Ground-Glass Opacity (GGO)的病變。傳統的腫瘤偵測演算法，通常是藉由一個物體的形態分割結果所擷取出的形態特徵來判斷此物體是否為肺腫瘤，更確切地說，它是根據一個特徵空間中形態特徵的空間分佈來做判斷。然而，這些方法的不足之處，就是難以定義一個適合的參數在分割的演算法上。如設定閥值產生物體邊界曲面的方法中，無法透過單一的閥值取得最佳描述物體形態特徵的邊界曲面；即使利用多重閥值的設定得到多組形態特徵的描述，可以改進這個方法的缺失，但因為在這之中不一定存在足以辨別的形態特徵，因此仍然有其潛在的爭議。
在分割參數的改變下所描述物體變異的可能性我們稱為形態特徵曲線，而形態動態的概念則是根據形態特徵曲線的圖形特徵來判定一個物體。我們提出的腫瘤偵測演算法，首先透過Frangi filtering將肺區中的每一個體素 (Voxel) 得到管狀的機率值，並將機率值高於設定參數之體素視為血管即可得到3D肺血管的重建。接著根據我們所提出之形態規模指數選擇出疑似腫瘤的位置。對於每一個位置，根據所得到的邊界曲面計算出五個形態特徵，並透過以決策樹為依據的分類器判斷是否為腫瘤。形態特徵曲線在本研究中為不同CT閥值下描述疑似腫瘤位置是否為腫瘤的一個函數，最後，每一個疑似腫瘤位置根據形態特徵曲線判斷是否為腫瘤。
我們所提出的3D腫瘤偵測演算法的效能評估上，是利用肺部圖像數據庫聯盟 (Lung Image Database Consortium, LIDC) 所提供的CT影像，每一張影像都經過四位放射科醫生的審查。我們使用LIDC數據庫的65組CT掃描當作我們的測試資料，每一組CT影像的剖面厚度為1.25毫米。我們只針對直徑等於或大於3毫米的腫瘤進行偵測，在至少一位放射科醫師的標註下，65組掃描總共包含76顆腫瘤。根據五倍的交叉驗證 (Five-fold Cross-validation) 方法下，對於肺區中附著於血管的腫瘤及懸浮於血管附近的腫瘤，我們提出的腫瘤偵測演算法達到86.8%的靈敏度及平均一組掃描1.52個誤判 (False-positives, FPs)。

The goal of this dissertation develop an effective CAD algorithm for lung nodule detection, the algorithm focuses on detection of lung nodules in the vascular trees within the lung, including the nodules attached to the vessels and those floating around the vessels. To achieve this goal, a new true 3D rule-based nodule detection algorithm is proposed in this dissertation. The uniqueness of the proposed detection algorithm lies in the novel idea of morphological dynamics.
To assist medical doctors in achieving a high lung nodule detection rate with a low false positive rate, varieties of computer-aided detection (CAD) algorithms have been developed previously, including several commercially available CAD systems and numerous laboratory CAD algorithms. However, none of these CAD algorithms have achieved satisfactory performances. Their common deficiencies are insufficient sensitivity, high false positive rate, and unable to detect GGO (Ground-Glass Opacity) lesions. Conventionally, to determine if an object of interest is a lung nodule is usually based on the morphological features extracted from the object shapes defined by one or several segmentation results. More precisely, it is based on the spatial distribution of the morphological features in the feature space. However, these approaches suffer a common deficiency. That is, it is hard to define the appropriate parameter values of the segmentation algorithms, e.g., the thresholding value, to derive the boundary surface best describing the characteristics of the object. Even if combining multiple sets of morphological features, each of which from a boundary surface defined by a set of parameter values, may remedy the deficiency, the potential advantages remain controversial because not all boundary surfaces may make contribution to the differentiation power.
The basic idea of morphological dynamics is to characterize the object based on the morphological characteristic curve, which describes the variation of the likelihood that an object of interest is a lung nodule as the parameter values change. In the proposed nodule detection algorithm, the lung vessels are first reconstructed by defining a voxel within the lung to be a vessel voxel if its vesselness is higher than a pre-specified thereshold, where the vesselness is calculated via Frangi filtering. The nodule candidates are then selected according to the morphological size index developed in this study. For each nodule candidate, five sets of morphological features are computed from the boundary surfaces, each derived by applying a thresholding CT number to the original lung CT images. For each set of morphological features, a nodule candidate is classified into either likely-to-be-nodule or likely-to-be-non-nodule through a rule-based decision tree. A morphological characteristic curve in this study is a function describing the likelihood of a nodule candidate being a nodule as the thresholding CT number changes. Finally, each nodule candidate is determined to be a nodule or a non-nodule based on the morphological characteristic curve.
The performance of the proposed 3D rule-based nodule detection algorithm is evaluated by using the lung CT images provided by the Lung Image Database Consortium (LIDC), each of which was reviewed by four radiologists. With a slice thickness of 1.25 mm, 65 scans containing 76 nodules are used in this study. Each nodule was marked as a nodule by at least one radiologist. All 76 nodules are larger than 3 mm. Based on the five-fold cross-validation method, for the nodules attached to the vessels and those floating around the vessels within the lung, the sensitivity of the proposed nodule detection algorithm is shown to be 86.6% with 1.52 false-positive nodules per scan.

論文目錄
第一章 緒論 1
1.1 研究背景 1
1.2 文獻回顧 4
第二章 理論與方法 9
2.1 數學形態學運算 9
2.2 肺部區域影像的分割方法 14
2.3 3D血管重建的方法 16
第三章 肺腫瘤CAD演算法 21
3.1 演算法架構流程圖 21
3.2 VOLUME OF INTEREST的選擇 23
3.3 形態動態法 27
3.4 腫瘤特徵萃取 35
3.4.1 第一階段腫瘤特徵萃取 35
3.4.2 第二階段腫瘤特徵萃取 37
3.5 分類器 38
第四章 實驗結果與討論 43
4.1 LIDC數據庫 43
4.2 實驗結果 46
4.2.1 懸浮腫瘤 46
4.2.2 附著腫瘤 49
4.3 結果分析與討論 58
第五章 結論與未來展望 62
參考文獻 63

參考文獻

[1] 行政院衛生署統計公布欄，http://www.doh.gov.tw/CHT2006/DM/DM2_2.aspx?class_no=440&level_no=1，2009。
[2] 柯獻欽，「頭號癌症殺手-肺癌」http://blog.udn.com/curecancer99/3107877，2009。
[3] 慈濟大學影像處理實驗室http://www.iplab.tcu.edu.tw/data/CT/CT_hi.htm，2007。
[4] Armato, III S. G., Giger, M. L., Moran, C. J., Blackburn, J. T., Doi, K., and MacMahon, H., “Computerized detection of pulmonary nodules on CT scans,” Radiographics, vol. 19, pp. 1303–1311, 1999.
[5] Armato, III S. G., Giger, M. L., and MacMahon, H., “Automated detection of lung nodules in CT scans: Preliminary results,” Med. Phys., vol. 28, pp. 1552–1561, 2001.
[6] Armato, III S. G., Altman, M. B., and LaRivière, P. J., “Automated detection of lung nodules in CT scans: Effect of image reconstruction algorithm,” Med. Phys. 30, 461–472, 2003.
[7] Bellotti, R., De Carlo, F., Gargano, G., Tangaro, S., Cascio, D., Catanzariti, E., Cerello, P., Cheran, S. C., Delogu, P., De Mitri, I., Fulcheri, C., Grosso, D., Retico, A., Squarcia, S., Tommasi, E., and Golosio, B., “A CAD system for nodule detection in low-dose lung CTs based on region growing and a new active contour model,” Med. Phys. 34, 4901–4910, 2007.
[8] Brown, M. S., Goldin, J. G., Suh, R. D., McNitt-Gray, M. F., Sayre, J. W., and Aberle, D. R., “Lung micronodules: automated method for detection at thin-section CT-initial experience,” Radiology 2003; 226:256-262.
[9] Dajnowiec, M., Alirezaie, J., and Paul, B., "An adaptive rule based automatic lung nodule detection system," in Pattern Recognition and Image Analysis, Lecture Notes in Computer Science, 2005, pp. 773-782.
[10] Das, M., Muhlenbruch, G., Mahnken, A. H., Flohr, T. G., Gundel, L., Stanzel, S., Kraus, T., Gunther, R. W., and Wildberger, J. E., “Small Pulmonary Nodules: Effect of Two Computer-aided Detection Systems on Radiologist Performance,” Radiology 2006; 241(2): 564-571.
[11] Erberich, S. G., Song, K., Arakawa, H., Huang, H. K., Richard, W., Hoo, K. S., and Loo, B.W., “Knowledge-based lung nodule detection from helical CT,” Radiology, vol. 205P, p. 617, 1997.
[12] Fischbach, F., Knollmann, F., Griesshaber, V., Freund, T., Akkol, E., and Felix, R., “Detection of pulmonary nodules by multislice computed tomography: improved detection rate with reduced slice thickness,” Eur Radiol 2003;13: 2378–2383.
[13] Frangi, A. F., Niessen, W. J., Vincken, K. L., and Viergever, M. A., “Multiscale vessel enhancement filtering,” In: Medical Image Computing and Computer-assisted Intervention (MICCAI’98), 1998, pp. 130–137.
[14] Gierada, D. S., Pilgram, T. K., Ford, M., Fagerstrom, R. M., Church, T. R., Nath, H., Garg, K., and Strollo, D. C., “Lung Cancer: Interobserver Agreement on Interpretation of Pulmonary Findings at Low-Dose CT Screening,” Radiology 2008; 246(1): 265-272.
[15] Giger, M. L., Bae, K. T., and MacMahon, H., “Computerized detection of pulmonary nodules in computed tomography images,” Investigat. Radiol.,vol. 29, pp. 459–465, 1994.
[16] Golosio, B., Masala, G. L., Piccioli, A., Oliva, P., Carpinelli, M., Cataldo, R., Cerello, P., De Carlo, F., Falaschi, F., Fantacci, M. E., Gargano, G., Kasae, P., and Torsello, M., “A novel multithreshold method for nodule detection in lung CT,” Med. Phys. 36, 3607–36182009.
[17] Gonzalez, R. C., and Woods, R. E., “Digital Image Processing 3rd ed,” ISBN 978-986-6534-10-2.
[18] Gori, I., Fantacci, M., Preite Martinez, A., and Retico, A., “An automated system for lung nodule detection in low-dose computed tomography,” Proceedings of the SPIE on Medical Imaging 2007: Computer-Aided Diagnosis, vol. 6514, p. 65143R. 2007.
[19] Gurcan, M. N., Sahiner, B., Petrick, N., Chan, H. P., Kazerooni, E. A., Cascade, P. N., and Hadjiiski, L., “Lung nodule detection on thoracic computed tomography images: Preliminary evaluation of a computer-aided diagnosis system,” Med. Phys. 29, 2552–2558, 2002.
[20] Hardie, R., Rogers, S., Wilson, T., and Rogers, A., “Performance analysis of a new computer aided detection system for identifying lung nodules on chest radiographs,” Medical Image Analysis 12 (3), 240–258. 2008.
[21] Heelan, R. T., Flehinger, B. J., Melamed, M. R., Zaman, M. B., Perchick, W. B., Caravelli, J. F., and Martini, N., “Non–small-cell lung cancer: results of the New York screening program,” Radiology 1984; 151:289–293.
[22] Henschke, C. I., Mccauley, D. I., Yankelevitz, D. F., Naidich, D. P., McGuinness, G., Miettinen, O. S., Libby, D., Pasmantier, M., Koizumi, J., Altorki, N., and Smith, J. P., “Early Lung Cancer Action Project: A Summary of the Findings on Baseline Screening,” The Oncologist 2001; 6: 147-152.
[23] Kanazawa, K., Kawata, Y., Niki, N., Satoh, H., Ohmatsu, H., Kakinuma, R., Kaneko, M., Moriyama, N., and Eguchi, K., “Computer-aided diagnosis for pulmonary nodules based on helical CT images,” Computerized Medical Imaging and Graphics 22 (2), 157–167. 1998.
[24] Lee, T.-C., Kashyap, R. L., and Chu, C. N., “Building skeleton models via 3-D medial surface/axis thinning algorithms,” Graph. Models Image Processing, vol. 56, no. 6, pp. 462–478, 1994.
[25] Lee, Y., Hara, T., Fujita, H., Itoh, S., and Ishigaki, T., “Automated detection of pulmonary nodules in helical CT images based on an improved template-matching technique,” IEEE Transactions on Medical Imaging 20 (7), 595–604. 2001.
[26] Manohar, M. and Ramapriyan, H.K., “Connected component labeling of binary images on a mesh connected massively parallel processor,” Comput. Vision, Graphics, and Image Process. 45 2 (1989), pp. 133–149.
[27] Matsumoto, S., and Kundel, H. L., “Pulmonary Nodule Detection In CT Images With Quantized Convergence Index Filter,” Medical Image Analysis, 2006, 10:343-352.
[28] McNitt-Gray, M. F., Armato, S. G. 3rd, Meyer, C. R., Reeves, A. P., McLennan, G., Pais, R. C., Freymann, J., Brown, M. S., Engelmann, R. M., Bland, P. H., Laderach, G. E., Piker, C., Guo, J., Twofic, Z., Qing D. P-Y., Yankelevitz, D. F., Aberle, D. R., van Beek, E. J., MacMahon, H., Kazerooni, E. A., Croft, B. Y., and Clarke, L. P., “The Lung Image Database Consortium (LIDC) data collection process for nodule detection and annotation,” Acad Radiol 2007;14:1464–1474.
[29] Messay, T., Hardie, R. C., and Rogers, S. K., “A new computationally efficient cad system for pulmonary nodule detection in CT imagery,” Med. Image Anal., vol. 14, no. 3, pp. 390–406, Jun. 2010.
[30] Opfer, R. and Wiemker, R., “Performance analysis for computer aided lung nodule detection on LIDC data,” Proc. SPIE 6515, 65151C.1–65151C.92007.
[31] Remy-Jardin, M., Remy, J., Giraud, F., and Marquette, C.H., “Pulmonary nodules: detection with thick-section spiral CT versus conventional CT,” Radiology 1993; 187:513–520.
[32] Retico, A., Delogu, P., Fantacci, M.E., Gori, I., and Preite Martinez, A., “Lung nodule detection in low-dose and thin-slice computed tomography,” Computers in Biology and Medicine 38 (4), 525–534. 2008.
[33] Rubin, G., Lyo, J., Paik, D., Sherbondy, A., Chow, L., Leung, A., Mindelzun, R.,Schraedley-Desmond, P., Zinck, S., Naidich, D., and Napel, S., “Pulmonary noduleson multi-detector row CT scans: performance comparison of radiologists and computer-aided detection,” Radiology 234 (1), 274. 2005.
[34] Saha, P. K., Chaudhuri, B. B., and Duuta Majumder, D., “A new shape preserving parallel thinning algorithm for 3D digital images,” Pattern Recogn 30 (1997), 1939–1955.
[35] Sahiner, B., Hadjiiski, L., Chan, H., Shi, J., Cascade, P., Kazerooni, E., Zhou, C., Wei, J., Chughtai, A., Poopat, C., Song, T., Nojkova, J. S., Frank, L., and Attili, A., “Effect of CAD on radiologists’ detection of lung nodules on thoracic CT scans: observer performance study,” Proceedings of SPIE 6515, 65151D. 2007.
[36] Sun, X., Zhang, H., and Duan, H., “3D computerized segmentation of lung volume with computed tomography,” Acad. Radiol. 13(6), 670–677 (2006).
[37] Suzuki, K., Armato, S.G., Li, F., Sone, S., and Doi, K., “Massive training artificial neural network (MTANN) for reduction of false positives in computerized detection of lung nodules in low-dose computed tomography,” Medical Physics 30 (7), 1602–1617. 2003.
[38] Yim, Y. and Hong, H., “Correction of segmented lung boundary for inclusion of pleural nodules and pulmonary vessels in chest CT images,” Comput. Biol. Med. 38 (8) (2008), pp. 845–857.
[39] Yuan, R., Vos, P. M., and Cooperberg, P. L., “Computer-Aided Detection in Screening CT for Pulmonary Nodules,” AJR 2006; 186:1280–1287.
[40] Zhao, B., Gamsu, G., Ginsberg, M. S., Jiang, L., and Schwartz, L. H., “Automatic detection of small lung nodules on CT utilizing a local density maximum algorithm,” J Appl Clin Med Phys 2003;4:248–260.