臺灣博碩士論文加值系統

肺癌為國人重要癌症死因，因此如何診斷肺癌成為近代醫學重要的研究課題。肺部電腦斷層掃描（CT，Computed Tomography）為診斷肺臟腫瘤最重要之工具。過去文獻顯示良、惡性病灶之顯影特性有所不同並代表著腫瘤血管之特徵。本研究使用動態顯影之CT影像，探討肺部結節影像特徵參數不同病灶之特性。
研究主要是對直徑小於3公分之單一肺部結節（Solitary Pulmonary Nodule）進行分析使用。共使用病案經證實之 82例（良性18例、惡性64例）動態顯影之CT影像；其顯影時間為顯影劑注射前（0秒）以及注射後25秒與50秒。利用影像處理之半自動分割技術，分別對包含腫瘤最大徑之單一切片與包含整體腫瘤之多張切片影像進行分割。使用紋理特徵含密度（Density）以及經由共相關矩陣（Co-correlation Matrix）所計算之熵（Entropy）、能量（Energy）、對比度（Contrast）和齊次性（Homogeneity），同時加上似圓性、面積/體積和直徑…等形態特徵進行參數分析。本研究使用之統計分析方法包含變異數分析以判斷顯影對於特徵參數之影響，t檢定分析以區別良、惡性腫瘤的差異，接收器操作特性曲線（Receiver Operating Characteristic）以曲線下面積（AUC，Area Under Curve）評估參數診斷效益；並利用倒傳遞類神經網路分析訓練組（良性11例、惡性39例）與測試組（良性7例、惡性25例）進行特徵參數之整合，以提升系統對良、惡性腫瘤的鑑別度。
研究結果顯示二維紋理參數於腫瘤最大徑之單一切片的統計結果大致優於腫瘤整體之多張切片合併的統計結果。以腫瘤最大徑之單一切片分析顯示，在惡性腫瘤之紋理參數Density（P=0.015）、Entropy（P=0.011）和Energy（P=0.024）受到顯影的影響改變較為顯著。在良、惡腫瘤群組間差異分析，Contrast於顯影前最為顯著（P=0.001）； Density（P=0.009）、Entropy（P=0.002）、Energy（P＜0.001）和Homogeneity（P=0.001）在顯影劑注射後50秒最為顯著。在診斷效益上，Contrast於顯影前所得之診斷效益最大（AUC=0.714），參數Density（AUC=0.725）、Energy（AUC=0.762）、Entropy（AUC=0.738）和Homogeneity（AUC=0.741）於顯影劑注射後50秒所得之診斷效益最大。透過倒傳遞類神經網路整合上述具診斷效益之參數，以隨機選取分成訓練與測試群組，可得到鑑別良、惡性準確率為0.791±0.017、靈敏度為0.908±0.018及專一度為0.457±0.107。
研究發現利用動態顯影之CT影像可偵測肺臟腫瘤於不同時間之顯影，因而改變其內在紋理特徵，使良、惡性肺臟病灶之區別更加明顯，同時提供紋理參數分析在影像臨床診斷之參考依據。未來可能使用多偵檢器電腦斷層掃描（MDCT，Multi-Detector Computed Tomography）儀進行體積掃描，以達縮短檢查時間以及提高影像解析度，幫助腫瘤血管新生特性之了解；若能配合電腦輔助診斷（Computer Aided Diagnosis）之技術處理其龐大的資料量，相信更可有效提升臨床診斷之效益。

Lung cancer is one of popular cancer for patient’s death in Taiwan. The diagnosis the lung cancer in early stage becomes very important research issue. Previous studies show that vascular enhancement character was different from the different kind of nodule in Computed tomography (CT) images. The purpose of this study is focused on feature analysis for early stage lung nodule in dynamic contrast enhancement CT image.
In this study, dynamic contrast enhanced CT images for lung nodule size smaller than 3 cm were used for analysis. Totally, 82 cases which had final histopathologic diagnosis by CT-guided biopsy and subsequent surgical biopsy as 18 benign cases and 64 malignant cases were used. Three sets of dynamic enhanced CT images that are 0 (pre-contrast), 25 (post-contrast1), and 50 (post-contrast2) seconds delay after contrast injection were acquired. Nodules within single slice image of the largest diameter of nodule or within multi-slice image of whole nodule were segmented thru a semi-automatic method. The characters of textural such as: density, entropy, energy, contrast, and homogeneity, and morphological parameters such as: area/volume, diameter, and circularity were extracted for feature analysis. Following by one way analysis of variance, independent sample t-test, and receiver operating characteristic (ROC) curve, the efficiency of system can be evaluated. Finally, we integrated the parameters with back propagation neural network (BPNN) to improve the power of nodule recognition.
Result shows that a single slice image can provided better information. From the comparison between different phases in malignant nodules CT images, significant differences were find in density (p=0.009), entropy (p=0.011), and energy (p=0.024). Comparison, There are significant differences between benign and malignant nodules of density (p=0.024), entropy (p=0.002), energy (p<0.001) and homogeneity (p=0.001) at post-contrast2 stage, and of contrast (p=0.001) at pre-contrast stage. The area under the ROC curve of density (0.725), entropy (0.762), energy (0.738) and homogeneity (0.741) are maximum at post-contrast2 stage, and of contrast (0.714) is maximum at pre-contrast stage. After repeated testing ten times, the accuracy, sensitivity, and specificity was 0.791±0.017, 0.908±0.018, 0.457±0.107, respectively for our system.
In conclusion, thru the comparison of 2D textural analyses between different stages of dynamic CT image, we find that dynamic contrast enhancement could change internal composition of lung nodule. This result leads useful parameters to classify benign and malignant nodules for further development in computer aided diagnosis (CAD) system. In the future, we expect that high spatial resolution multi-detector computed tomography which creates a large amount data are used to observe that vascular enhancement character in CT image, and then diagnostic efficacy need to be evaluated by match up technique of CAD.

目錄
摘要 I
Abstract III
謝誌 V
目錄 VI
圖索引 VIII
表索引 X
第一章、緒論 1
1-1前言 1
1-2研究背景 1
1-3文獻回顧 4
1-4研究目的 6
1-5論文架構 7
第二章、理論基礎 8
2-1肺部電腦斷層影像 8
2-1-1肺部電腦斷層掃描 8
2-1-2 X光用正影對比劑 10
2-1-3 DICOM標準簡述 11
2-2肺癌放射特徵與鑑別診斷 13
2-3影像處理技術 14
2-3-1影像描述統計 14
2-3-2臨界值法 15
2-3-3影像參數擷取 16
2-3-3.1紋理參數 16
2-3-3.2形態參數 18
2-4統計分析 19
2-4-1單因子變數分析 19
2-4-2獨立樣本t檢定 22
2-4-3診斷統計學 23
2-4-3.1決策矩陣 23
2-4-3.2接收器操作特性曲線 25
2-5 倒傳遞類神經網路 26
第三章、研究架構及方法 28
3-1材料與設備 28
3-2研究設備 30
3-3研究流程 30
3-4研究方法 32
3-4-1影像處理 32
3-4-2參數分析 35
第四章、結果與討論 37
4-1系統準確性評估 37
4-1-1 DICOM檔影像讀取 37
4-1-2 ROI遮罩之評估 39
4-2影像資料的選擇 40
4-2-1排除之病例 40
4-2-2病因統計 41
4-3參數擷取 41
4-3-1顯影時間序 42
4-3-2腫瘤ROI選取 46
4-3-2.1腫瘤內部ROI之分析結果 46
4-3-2.2完整腫瘤ROI之分析結果 47
4-3-3 CT系列影像參數擷取 48
4-3-4單一切片之T檢定結果 49
4-3-5多張切片之T檢定結果 50
4-3-6 ROC曲線分析之結果 52
4-3-7倒傳遞類神經網路 56
4-4腫瘤性質之影像參數討論 56
第五章、結論與未來展望 60
5-1結論 60
5-2未來展望 61
參考文獻 62
附錄A 單一切片腫瘤內部ROI之One Way ANOVA 67
附錄B 單一切片完整腫瘤ROI之One Way ANOVA 69
附錄C 多張切片完整腫瘤ROI之One Way ANOVA 72
附錄D 單一切片之Independent Sample t-Test 75
附錄E 多張切片之Independent Sample t-Test 80
附錄F 各參數之ROC曲線下面積(AUC) 82
作者自述 84

圖索引
圖1-1 肺部結節之CT影像 3
圖2-1 螺旋電腦斷層掃描示意圖 9
圖2-2 DICOM 檔案格式示意圖 12
圖2-3 資料元件結構圖 12
圖2-4 閥值擷取示意圖 15
圖2-5 紋理參數示意圖 17
圖2-6 似圓性圈選示意圖 18
圖2-7 voxel 之示意圖 18
圖2-8 檢查結果之分布情況圖 24
圖2-9各臨界值診斷曲線圖 25
圖2-10 ROC曲線圖 26
圖3-1 顯影劑注射之方式 29
圖3-2 動態顯影之肺部CT影像 30
圖3-3 研究流程圖 31
圖3-4 Ultra-Edit32開啟之DICOM影像 32
圖3-5 半自動分割CT影像 33
圖3-6 顯影時序圖 34
圖3-7 不同ROI圈選示意圖 35
圖3-8 切片影像擷取示意圖； 35
圖3-9 倒傳遞類神經網路架構 36
圖4-1 TomoVision DICOM 軟體介面圖 38
圖4-2 tag資訊驗證示意圖 39
圖4-3 假體展示圖 40
圖4-4 中心空洞之腫瘤影像 40
圖4-5 惡性腫瘤範例於時間序列上之影像表現 43
圖4-6 惡性腫瘤範例之參數分析圖 44
圖4-7 良性腫瘤範例於時間序列上之影像表現 44
圖4-8 良性腫瘤範例之參數分析圖 45
圖4-9 不同ROI圈選示意圖 46
圖4-10 惡性腫瘤於三個顯影時間點之影像 48
圖4-11 Density於惡性腫瘤case 1.之動態顯影剖繪圖 55
圖4-12 Density於惡性腫瘤case 2.之動態顯影剖繪圖 55
圖4-13 Density於顯影後50s，各子群組間之t-test 58
圖4-14 Entropy於顯影後50s，各子群組間之t-test 58
圖4-15 Energy於顯影後50s，各子群組間之t-test 58
圖4-16 Homogeneity於顯影後50s，各子群組間之t-test 59
圖4-17 Contrast於顯影前，各子群組間之t-test 59
圖4-18 Difference of Density 25秒到0秒，各子群組間之t-test 59

表索引
表2-1 圖2-5之紋理參數計算表 17
表2-2 單因子分類資料表 20
表2-3 變異數分析表 20
表2-4 決策矩陣 24
表3-1 電腦斷層儀掃瞄參數表 28
表4-1 結節診斷統計表 41
表4-2 單一切片之Independent Sample t-Test結果 51
表4-3 多張切片之Independent Sample t-Test結果 53
表4-4 紋理參數篩檢值對應表 54
表4-5 子群組比較對照表 57
表A-1 良性參數之變異數分析表 67
表A-2 惡性參數之變異數分析表 67
表A-3 惡性參數之事後檢驗表 68
表B-1 良性參數之變異數分析表 69
表B-2 惡性參數之變異數分析表 70
表B-3 惡性參數之事後檢驗表 71
表C-1 良性參數之變異數分析表 72
表C-2 惡性參數之變異數分析表 73
表C-3 惡性參數之事後檢驗表 74
表D-1 單一切片之紋理參數於描述性統計表 75
表D-2 單一切片之Independent Sample t-Test分析表 78
表E-1 多張切片之紋理參數於描述性統計表 80
表E-2 多張切片之Independent Sample t-Test分析表 81
表F-1 單一切片擷取各參數之AUC 82
表F-2 多張切片擷取各參數之AUC 83

[1] 行政院衛生署網站，http://www.doh.gov.tw，民國九十五年，六月
[2] S. Momoshima著作，黃明賢 譯，“圖解胸部X-光片”，長年出版社，pp.85-121，民國八十九年
[3] 謝博生、林肇堂、楊泮池、江福田，“臨床內科學 疾病篇(上冊) ”，金名圖書有限公司，pp.1084-1087，民國八十二年
[4] 國家衛生研究院癌症研究組臺灣癌症臨床研究合作組織，“肺癌診治共識”，國家衛生研究院，民國八十七年，八月
[5] K. Matsuyama, Y. Chiba, M. Sasaki, H. Tanaka, R. Muraoka, and N. Tanigawa, “Tumor Angiogenesis as a Prognostic Marker in Operable Non–Small Cell Lung Cancer,” The Annals of Thoracic Surgery, vol.65 ,pp.1405-1409, 1998
[6] N. Tanigawa, M. Matsumura, H. Amaya, A. Kitaoka, T. Shimomatsuya, C. Lu et al. “Tumor vascularity correlates with the prognosis of patients with esophageal squamous cell carcinoma,” Cancer, vol.79, pp.220-225, 2000
[7] K. Fujimoto, T. Abe, N.L. Müller, H. Terasaki, S. Kato, J. Sadohara et al. “Small Peripheral Pulmonary Carcinomas Evaluated with Dynamic MR Imaging: Correlation with Tumor Vascularity and Prognosis,” Radiology, vol.227, pp.786-793, 2003
[8] P. Armstrong and M.L. Wastie著作，邱健泰、王仲祺、黃忠英、林怡年 譯，“影像診斷學”，藝軒圖書出版社，pp.18-21，民國八十三年
[9] 陸坤泰與張登斌，“台灣的肺癌”，行政院衛生署八十年度委託研究計畫研究報告
[10] S.J. Swensen, R.L. Morin, B.A. Schueler, L.R. Brown, D.A. Cortese, P.C. Pairolero et al., “Solitary Pulmonary Nodule: CT Evaluation of Enhancement with Iodinated Contrast Material－A Preliminary Report,” Radiology, vol.182, pp.343-347, 1992
[11] S.J. Swensen, R.W. Viggiano, D.E. Midthun, N.L. Müller, A. Sherrick, K. Yamashita et al., “Lung Nodule Enhancement at CT: Multicenter Study,” Radiology, vol.214, pp.73-80, 2000
[12] J.J. Erasmus, H.P. McAdams, and J.E. Connolly, “Solitary Pulmonary Nodules: Part II. Evaluation of the Indeterminate Nodule,” RadioGraphics, vol.20, pp.59-66, 2000
[13] K. Minami, Y. Kawata, N. Niki, K. Moril, H. Ohmatsu, R. Kakinuma et al., “Computerized characterization of contrast enhancement patterns for classifying pulmonary nodules,” International Conference on Image Processing, vol.2, pp.897-900, 2001
[14] C.A. Yi, K.S. Lee, E.A. Kim, J. Han, H. Kim, O.J. Kwon et al., “Solitary Pulmonary Nodules: Dynamic Enhanced Multi–Detector Row CT Studyand Comparison with Vascular Endothelial Growth Factor and Microvessel Density,” Radiology vol. 233, pp191–199, 2004
[15] A.V. Proto and S.R. Thomas, “Pulmonary nodules studied by computed Tomography,” Radiology, vol. 156, pp149-153, 1985
[16] T. Tozaki, Y. Kawata, N. Niki, H. Ohmatsu, R. Kakinuma, K. Eguchi, et al., “Pulmonary organs analysis for differential diagnosis based on thoracic thin-section CT images,” IEEE Trans. on Nuclear science, vol.45, pp.3075-3082, 1998
[17] N. Takagi, Y. Kawata, N. Niki, K. Mori, H. Ohmatsu, R. Kakinuma, et al., “3D analysis of solitary pulmonary nodules based on contrast enhanced dynamic CT,” International Conference on Image Processing, vol.3, pp.416-420, 1999
[18] Y. Kawata, N. Niki, H. Ohmatsu, M. Kusumoto, R. Kakinuma, K. Mori et al., “Computerized analysis of pulmonary nodules in topological and histogram feature spaces,” 15th International Conference on Pattern Recognition, vol.4, pp.332-335, 2000
[19] Y. Kawata, N. Niki, H. Ohmatsu, M. Kusumoto, R. Kakinuma, K. Mori et al., “Internal structure analysis of pulmonary nodules in topological and histogram feature spaces,” International Conference on Image Processing, vol.1, pp.168-171, 2000
[20] Y. Kawata, N. Niki, H. Ohmatsu, R. Kakinuma, K. Eguchi, M. Kaneko et al., “Quantitative surface characterization pulmonary nodules based on thin-section CT image,” IEEE Trans. Nuclear Science, vol.45, pp.2132-2138, 1998
[21] Y. Kawata, N. Niki, H. Ohmatsu, M. Kusumoto, R. Kakinuma, K. Mori et al., “Surrounding structures analysis of pulmonary nodules using differential geometry based vector fields,” International Conference on Image Processing, vol.3, pp.424-427, 2000
[22] M.F. McNitt-Gray, E.M. Hart, N. Wyckoff, J.W. Sayre, J.G. Goldin, and D.R. Aberle, “A pattern classification approach to characterizing solitary pulmonary nodules imaged on high resolution CT: Preliminary results,” Medical Physics, vol.26, pp.880-888, 1999
[23] 王峻峰，“單發性結節的智慧型形態鑑別方法”，中山大學機械工程研究所碩士論文，民國八十八年
[24] 莊美恩，“影像處理在肺部腫瘤偵測之應用”，中原大學醫學工程研究所碩士論文，民國八十九年
[25] 陳逸雯，“肺部腫瘤偵測之電腦輔助診斷系統”，中原大學醫學工程研究所碩士論文，民國九十一年
[26] C.F. Yeh, H.H. Cheng, Y.C. Chang, and J.L. Su, “The 3D morphological analysis method for pulmonary nodule diagnosis,” Chung Yuan Journal, vol. 32, pp.363-372, 2004.
[27] 楊順聰，“臨床高等醫療儀器系統-原理與應用”，藝軒圖書出版社，民國八十七年，三月
[28] S. Jackson and R. Thomas, “Cross-sectional Imaging Made Easy,” Elsevier, pp.5-7, 2004
[29] 莊克士，“醫學影像物理學”，合記圖書出版社，pp.411-414，pp.238-244，pp.475-483，民國九十二年，十月
[30] D.A. Clunie, “DICOM Structured Reporting,” PixelMed Publishing, 2000
[31] DICOM 3.0, “Digital Imaging and Communications in Medicine(DICOM)- Part 3:Information Object Definitions,” ACR-NEMA, 2004
[32] DICOM 3.0, “Digital Imaging and Communications in Medicine(DICOM)- Part 5: Data Structures and Encoding,” ACR-NEMA, 2004
[33] 繆紹綱，“數位影像處理 活用-Matlab”，全華科技圖書股份有限公司，p8-2－8-6，民國九十三年，四月
[34] C.M. Wu, Y.C. Chen, and K.S. Hsieh, “Texture features for classification of ultrasonic liver images,” IEEE Trans. on Medical Imaging , vol.11, pp.141-152, 1992
[35] E.L. Chen, P.C. Chung, C.L. Chen, H.M. Tsai, and C.I. Chang, “An automatic diagnostic system for CT liver image classification,” IEEE Trans. on Biomedical Engineering, vol.45, pp.783-794, 1998
[36] R.C. Gonzalez and R.E. Woods, “Digital image process,” 2nd ed., Prentice-Hall, Upper Saddle River, New Jersey, pp.665-669, 2002
[37] A.P. Dhawan, “Medical Image Analysis,” John Wiley & Sons., Inc., Hoboken, New Jersey, pp.59-63, 2003
[38]沈明來，“試驗設計學”，九州圖書，pp.31-46，pp.67-84，民國九十三年，二月
[39] W.W. Daniel, “Biostatistics: a foundation of analysis in health sciences,” 6th ed., John Wiley & Sons, Toronto, Ontario, Canada, pp. 252-255, 1999
[40] A.N. Esgiar, R.N.G. Naguib, B.S. Sharif, M.K. Bennett, and A. Murray, “Microscopic image analysis for quantitative measurement and feature identification of normal and cancerous colonic mucosa,” IEEE Trans. on Information Technology in Biomedicine, vol.2, pp.197-203, 1998
[41] S. Herlidou-Meme, J.M. Constans, B. Carsin, D. Olivie, P.A. Eliat, L. Nadal-Desbarats et al., “MRI texture analysis on texture test objects, normal brain and intracranial tumors,” Magnetic Resonance Imaging , vol.21, pp.989-993, 2003
[42] 王進德、蕭大全，“類神經網路與模糊控制理論入門”，全華圖書，民國九十二年，十月
[43] A.C. Fleischer, “Sonographic depiction of tumor vascularity and flow: from in vivo models to clinical applications,” Journal of Ultrasound in Medicine, vol.19, pp.55-61, 2000
[44] T. Hosoki, Y. Yosioka, T. Matsubara, K. Minamitani, M. Higashi, M. Ohtani et al., “Power Doppler sonography of hepatocellular carcinoma treated by transcatheter arterial chemoembolization. Assessment of the therapeutic effect,” Acta Radiologica, vol.40, pp.639-643, 1999