成功大學博碩士論文系統

簡易檢索 / 詳目顯示

回結果列表

研究生：	鄭子琪 Cheng, Tsu-Chi
論文名稱：	以機器學習開發氟化去氧葡萄糖正子造影電腦斷層影像的非小細胞肺癌淋巴結轉移診斷系統 Development of lymph node metastasis diagnosis system for patients with non-small-cell lung cancer (NSCLC) on F-18-FDG PET/CT images via machine learning algorithm
指導教授：	方佑華 Fang, Yu-Hua
學位類別：	碩士 Master
系所名稱：	工學院 - 生物醫學工程學系 Department of BioMedical Engineering
論文出版年：	2019
畢業學年度：	107
語文別：	英文
論文頁數：	72
中文關鍵詞：	非小細胞肺癌、F-FDG-PET/ CT 、淋巴結轉移、淋巴引流通路、機器學習、分類樹、集合學習
外文關鍵詞：	Non-small cell lung cancer, F-FDG-PET/CT, Lymph node, Lymphatic drainage pathway, Machine learning, Decision tree, Ensemble learning, RusBoost, Metastasis
相關次數：	點閱：87 下載：0
分享至:	分享至facebook 分享至twitter

查詢本校圖書館目錄查詢臺灣博碩士論文知識加值系統勘誤回報

肺癌是全球死亡的主要原因，約有85％肺癌為非小細胞肺癌（NSCLC）。對於非小細胞癌患者而言，淋巴結轉移的分期極為重要，因為這關係到其治療方法及預後。對於非小細胞肺癌而言在影像診斷上，相較於其他非侵入性的影像診斷，FDG-PET/CT影像在影像診斷中淋巴結分期中的具有較佳之準確性。在影像中如果SUV 越大代表惡性的機率越大，因為腫瘤細胞會攝取FDG使得SUV值增加。然而在產生發炎反應的組織中也會有產生攝取過多FDG 的現象發生。因此FDG-PET/CT影像在非小細胞癌淋巴結轉移中仍然需要改進。由過去的研究中我們可以得知除了FDG-PET/CT的傳統圖像參數，如: 標準攝取值（〖SUV〗_max），尚有非常多參數可以由圖像中獲取，好比淋巴引流通路以及其他從影像上獲取之紋理特徵。
本論文獲取成大醫院人體研究計畫案依回溯性研究的方式納入從2014年12月日至2018年12月之123位病患，其包含576顆疑似惡性淋巴結。在所有淋巴結中，43顆之病理切片判讀為惡性之轉移淋巴結。所有病患皆經過早期相及延遲相之PET/CT 檢查，並具有原發腫瘤，淋巴結之位置及病理報告結果。為了提供臨床醫師一些由FDG-PET/CT影像中診斷出非小細胞癌患者的淋巴結轉移的指引，本研究開發了一種電腦輔助診斷（CAD）系統，利用機器學習演算法中的集成學習的方式來提高診斷效率。至於研究中數據不平衡的問題，本研究之解決方案是使用RusBoost進行集成學習。最終在此研究中之模型包括四個特徵，即在路徑最末端之結點，延遲相的〖SUV〗_max，原發腫瘤的〖SUV〗_mean,紋理特徵中的相素中之長期強調。通過機器學習算法中的集成學習提高診斷效率，本次研究之結果為: 91.23％的準確率，90％的靈敏度和92.5％的特異性。由本次實驗之結果可以得知比起單參數多參數對於非小細胞肺癌之診斷更加準確,並且針對這類不平衡的數據的醫學影像分析集成學習的效果的會更加顯著。在臨床上電腦輔助診斷系統之判斷結果能夠協助醫師的診斷，使得臨床醫學影像診斷能因其而獲取更好的準確性。

Lung Cancer is a serious health threat worldwide, it can be separated into two kinds of lung cancer. About 85% of lung cancer patients suffer from non-small cell lung cancer, which is abbreviation as NSCLC. The staging of lymph nodes in NSCLC patients is extremely important because respective stages require different treatments. For non-small cell lung cancer, 18F-2-fluoro-2-deoxy-d-glucose (FDG) PET/CT images have a better accuracy in lymph node staging comparing with the non-invasive images. However, the results of discriminating lymph node staging on FDG positron emission tomography (PET) / computed tomography (CT) still needs improvement. In addition to the traditional image parameters of FDG-PET/CT such as standardized uptake value (SUV), there are many other parameters available from FDG-PET/CT images, for example, the lymphatic drainage pathway. In this thesis, with the approval of the Institutional Review Board, we included 123 patients with a total of 576 lymph nodes, among which were and 43 malignant. All patients underwent early phase scan and delay phase scan in PET/CT scanner and the information we included the primary tumor and lymph nodes pathological reports and locations of them.
For the purpose of providing a guide for clinical physicians to diagnosis on lymph node metastasis on NSCLC patients with FDG-PET/CT image, this research developed a computer-aided diagnosis (CAD) system by using a machine learning algorithm. As for the problem of imbalanced data in research, our solution was the use of ensemble learning with RusBoost algorithm. Our final model included four features, including the leaf status of the node, 〖SUV〗_max in delay the phase, 〖SUV〗_mean of the primary tumor and the long run emphasis of voxel-alignment. By using the ensemble learning to improve the diagnostic efficiency, our research achieved an accuracy of 91.23%, a sensitivity of 90% and a specificity of 92.5%. From the results of this experiment, it showed that the diagnosis of non-small cell lung cancer is more accurate than the single parameter multi-parameter, and ensemble learning is more significant in the imbalanced data. The results of the computer-aided diagnosis system can help clinical physicians with their diagnosis, which can obtain a better accuracy in diagnostic.

摘 要	i
Abstract	iii
Table of Contents	vii
List of Tables	x
List of Figures	xi
Chapter 1 Overview	1
1.	Overview	1
Chapter 2 Introduction	3
1	Lung Cancer	3
1.1	The Importance of Determining Lymph Node Metastasis in NSCLC	4
1.2	The Principle and Clinical Use of F-18 FDG PET	7
1.3	F-18-FDG-PET/CT	8
1.4	The Shortage of Determining Lymph Node Metastasis by F-18-FDG-PET/CT	9
2	Computer Aided Diagnosis (CAD) System	10
2.1	The Benefit of the CAD System	10
2.2	Image Processing Technology	11
2.2.1	The Maximum Standardized uptake value	11
2.2.2	Dual Time Image	12
2.2.3	Primary Tumor SUVmax and Size	13
2.2.4	Nodal Size	14
2.2.5	Lymphatic Drainage Pathways (LDP)	15
2.3	Texture Analysis	17
2.4	The Current CAD System	19
3	Machine Learning	20
3.1	Support Vector Machines	21
3.2	Classification Tree	22
3.3	Imbalanced Data	23
3.4	Ensemble Learning	24
4	Specific Aim	25
Chapter 3 Materials and Methods	28
1	Data Acquisition	29
2	Data Preprocessing and Preparation	29
2.1	The Segmentation Software	29
2.2	The Definition of the Size of Region of Interest	30
2.3	Quantitative Image Parameters	31
3	Data Classifier	35
3.1	Interpretation of Support Vector machine	35
3.2	Interpretation of Classification Tree Classifier	35
3.3	Imbalanced Classes in Classification Tree Classifier	36
3.4	Ensemble Learning	36
3.5	Boosting Algorithms	37
3.5.1	AdaBoost	38
3.5.2	RUSBoost	39
4	Evaluation Methods	41
4.1	Accuracy, Sensitivity, and Specificity	41
4.2	The Receiver Operating Characteristics (ROC)Curve and Area Under the Curve (AUC)Value	43
5	Validation Technique	44
5.1	k -fold Cross Validation (kfoldCV)	45
5.2	Leave-One-Out Cross Validation (LOOCV)	46
Chapter 4 Results	47
1	Classification in Single Parameter	47
2	The Analysis with Multi Parameters	49
3.1	The Analysis with Different Classifiers	50
3.2	The Analysis with Different Boosting Algorithm	52
3.3	The Analysis with Different number of features	53
Chapter 5 Discussion and Future Work	56
1	The Analysis of Nodal Diagnosis	56
2	The Impact of Feature	57
3	Comparison Result with the Existing Literatures	60
4	Future Work	62
References	65

                                    

[1] R. L.Siegel, K. D.Miller, and A.Jemal, “Cancer Statistics , 2017,” vol. 67, no. 1, pp. 7–30, 2017.
[2] R. M.Patricia, D.Frank, and M. A.C, “Diagnosis of Lung Cancer *,” Chest, vol. 123, no. 1, pp. 129S-136S, 2003.
[3] C. K.Hoh, “Clinical use of FDG PET,” vol. 34, pp. 737–742, 2007.
[4] F.Bray, J.Ferlay, and I.Soerjomataram, “Global Cancer Statistics 2018 : GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries,” pp. 394–424, 2018.
[5] S.Tumors, “Non–Small Cell Lung Cancer: Epidemiology, Risk Factors, Treatment, and Survivorship,” vol. 83, no. May, pp. 584–594, 2008.
[6] A.Risch and C.Plass, “Lung cancer epigenetics and genetics,” vol. 7, no. March, pp. 1–7, 2008.
[7] T.Canadian and L.Oncology, “Investigation for Mediastinal Disease in Patients With Apparently Operable Lung Cancer,” vol. 4975, no. 1.
[8] C. A. B.Saunders, J. E.Dussek, M. J. O.Doherty, and M. N.Maisey, “Evaluation of Fluorine-18-Fluorodeoxyglucose Whole Body Positron Emission Tomography Imaging in the Staging of Lung Cancer,” vol. 4975, no. 99, 1999.
[9] C. F.Mountain, “special report Revisions in the International System for Staging Lung Cancer *,” Chest, vol. 111, no. 6, pp. 1710–1717, 1997.
[10] C.Investigations, “Mediastinal Lymph Node Staging by FDG-PET in Patients with Non-Small Cell Lung Cancer : Analysis of False-Positive FDG-PET Findings,” vol. 8638, pp. 500–506, 2003.
[11] D. C.Mccrory, “Noninvasive Staging of Non-small Cell Lung Cancer *,” Chest, vol. 123, no. 1, pp. 137S-146S, 2003.
[12] A. E. C.Practice, “Noninvasive Staging of Non-small Cell Lung Cancer *,” Chest, vol. 132, no. 3, pp. 178S-201S, 2007.
[13] A. E.Clinical and P.Guidelines, “Invasive Mediastinal Staging of Lung Cancer *,” Chest, vol. 132, no. 3, pp. 202S-220S, 2007.
[14] T.Schroeder et al., “Detection of Pulmonary Nodules Using a 2D HASTE MR Sequence : Comparison with MDCT,” no. October, pp. 979–984, 2005.
[15] R.Kono, K.Fujimoto, H.Terasaki, N. L.Müller, S.Kato, and S.Takamori, “Dynamic MRI of Solitary Pulmonary Nodules : Comparison of Enhancement Small Peripheral Lung Lesions,” no. January, pp. 26–36, 2007.
[16] E. M.Kamel et al., “Staging of Non–Small-Cell Lung Cancer with Integrated Positron-Emission Tomography and Computed Tomography,” pp. 2500–2507, 2003.
[17] Kazuhiro Yasufuku,“Comparison of Endobronchial Ultrasound, Positron Emission Tomography, and CT for Lymph Node Staging of Lung Cancer,” pp. 1–9, 2019.
[18] A.Fritscher-Ravens et al., “Mediastinal lymph node involvement in potentially resectable lung cancer: Comparison of CT, positron emission tomography, and endoscopic ultrasonography with and without fine-needle aspiration,” Chest, vol. 123, no. 2, pp. 442–451, 2003.
[19] K. A.McLaughlin et al., “PET, CT, and MRI with Combidex for mediastinal staging in non-small cell lung carcinoma,” Ann. Thorac. Surg., vol. 68, no. 3, pp. 1022–1028, 2002.
[20] J. T.Annema, O. S.Hoekstra, E. F.Smit, M.Veseliç, M. I. M.Versteegh, and K. F.Rabe, “Towards a minimally invasive staging strategy in NSCLC: Analysis of PET positive mediastinal lesions by EUS-FNA,” Lung Cancer, vol. 44, no. 1, pp. 53–60, 2004.
[21] S.Tsim, C. A.O’Dowd, R.Milroy, and S.Davidson, “Staging of non-small cell lung cancer (NSCLC): A review,” Respir. Med., vol. 104, no. 12, pp. 1767–1774, 2010.
[22] A.Rosset, L.Spadola, and O.Ratib, “OsiriX: An Open-Source Software for Navigating in Multidimensional DICOM Images,” vol. 17, no. 3, pp. 205–216, 2004.
[23] K.Doi, “Computer-aided diagnosis in medical imaging : Historical review, current status, and future potential,” vol. 31, pp. 198–211, 2007.
[24] H.Wang et al., “Comparison of machine learning methods for classifying mediastinal lymph node metastasis of non-small cell lung cancer from 18 F-FDG PET / CT images,” EJNMMI Res., 2017.
[25] D.Hellwig, T. P.Graeter, D.Ukena, A.Groeschel, and G. W.Sybrecht, “PET for Mediastinal Staging of Lung Cancer : Which SUV Threshold Makes Sense ?,” pp. 1761–1767, 2019.
[26] H.Zhuang et al., “Dual Time Point 18 F-FDG PET Imaging for Differentiating Malignant from Inflammatory Processes,” vol. 42, no. 9, pp. 1412–1418, 2019.
[27] L. M.Hamberg, G. J.Hunter, N. M.Alpert, N. C.Choi, J. W.Babich, and A. J.Fischman, “The dose uptake ratio as an index of glucose metabolism: useful parameter or oversimplification?,” J. Nucl. Med., vol. 35, no. 8, pp. 1308–12, 1994.
[28] M. A.Lodge, J. D.Lucas, P. K.Marsden, B. F.Cronin, M. J. O.Doherty, and M. A.Smith, “Original article A PET study of 18 FDG uptake in soft tissue masses,” Eur. J. Nucl. Med., vol. 26, no. 1, 1999.
[29] M.Li et al., “Primary tumor PET/CT [18F]FDG uptake is an independent predictive factor for regional lymph node metastasis in patients with non-small cell lung cancer,” Cancer Imaging, vol. 12, no. 3, pp. 566–572, 2012.
[30] L.Li et al., “Risk factors for predicting the occult nodal metastasis in T1-2N0M0NSCLC patients staged by PET/CT: Potential value in the clinic,” Lung Cancer, vol. 81, no. 2, pp. 213–217, 2013.
[31] A. J.DeLangen, P.Raijmakers, I.Riphagen, M. A.Paul, and O. S.Hoekstra, “The size of mediastinal lymph nodes and its relation with metastatic involvement : a meta-analysis,” vol. 29, pp. 26–29, 2006.
[32] S. L.Loomis, M. S. K. H.Taber, and L.Suzanne, “Patterns of Lymphadenopathy in,” Radiographics, pp. 419–434, 2004.
[33] A.Taghizadeh Kermani, R.Bagheri, S.Tehranian, P.Shojaee, R.Sadeghi, andD.N. Krag, “Accuracy of sentinel node biopsy in the staging of non-small cell lung carcinomas: Systematic review and meta-analysis of the literature,” Lung Cancer, vol. 80, no. 1, pp. 5–14, 2013.
[34] A. H.El-Sherief, C. T.Lau, C. C.Wu, R. L.Drake, G. F.Abbott, and T. W.Rice, “International Association for the Study of Lung Cancer (IASLC) Lymph Node Map: Radiologic Review with CT Illustration,” RadioGraphics, vol. 34, no. 6, pp. 1680–1691, 2014.
[35] Y.Shigemoto, K.Suga, and N.Matsunaga, “F-18-FDG-avid lymph node metastasis along preferential lymphatic drainage pathways from the tumor-bearing lung lobe on F-18-FDG PET/CT in patients with non-small-cell lung cancer,” Ann. Nucl. Med., vol. 30, no. 4, pp. 287–297, 2016.
[36] F.Tixier et al., “Concomitant Radiochemotherapy in Esophageal Cancer,” pp. 369–379, 2019.
[37] S.Chicklore et al., “Quantifying tumour heterogeneity in imaging by texture analysis,” pp. 133–140, 2013.
[38] J. F.Veenland, J. L.Grashuis, and E. S.Gelsema, “Texture analysis in radiographs: The influence of modulation transfer function and noise on the discriminative ability of texture features,” pp. 922–936, 1998.
[39] F.Tixier et al., “Reproducibility of Tumor Uptake Heterogeneity,” pp. 693–701, 2019.
[40] Y.Nishiyama, Y.Yamamoto, N.Kimura, S.Ishikawa, Y.Sasakawa, and M.Ohkawa, “Dual-time-point FDG-PET for evaluation of lymph node metastasis in patients with non-small-cell lung cancer,” Ann. Nucl. Med., vol. 22, no. 4, pp. 245–250, 2008.
[41] “Artificial Intelligence and Machine Learning : Policy Paper,” no. April, 2017.
[42] M. N.Wernick, Y.Yang, J. G.Brankov, G.Yourganov, and S. C.Strother, “Drawing conclusions from medical images ],” no. July, pp. 25–38, 2010.
[43] B.Zhang et al., “Radiomic machine-learning classifiers for prognostic biomarkers of advanced nasopharyngeal carcinoma,” Cancer Lett., vol. 403, pp. 21–27, 2017.
[44] S.Tong andE.Chang, “Support Vector Machine Active Learning for Image Retrieval,” pp. 107–118, 2001.
[45] T.Joachims, “Making large-scale SVM learning practical,” 1998.
[46] Robert M.Haralick,“The table look-up rule.”1976 ..
[47] I.Introduction, “A Survey of Decision Wee Classifier Methodology,” vol. 21, no. 3, 1991.
[48] Y.Liu and X.Yao, “Ensemble learning via negative correlation,” vol. 12, pp. 1399–1404, 1999.
[49] J. M.Tomczak, M.Lubicz, J.Swi, and M.Zi, “Boosted SVM for extracting rules from imbalanced data in application to prediction of the post-operative life expectancy in the lung cancer patients,” vol. 14, pp. 99–108, 2014.
[50] Y.Sui, Y.Wei, and D.Zhao, “Computer-Aided Lung Nodule Recognition by SVM Classifier Based on Combination of Random Undersampling and SMOTE,” vol. 2015, 2015.
[51] A.Teramoto and H.Fujita, “Automated detection of pulmonary nodules in PET / CT images : Ensemble false-positive reduction using a convolutional neural network technique,” vol. 43, no. 6, pp. 2821–2827, 2016.
[52] R.Lawrence, A.Bunn, S.Powell, and M.Zambon, “Classification of remotely sensed imagery using stochastic gradient boosting as a refinement of classification tree analysis,” vol. 90, pp. 331–336, 2004.
[53] A. H.Strahler, “The Use of Prior Probabilities in Maximum Likelihood Classification of Remotely Sensed Data,” vol. 163, pp. 135–163, 1980.
[54] G.Lemaˆ and M.Radojevi, “Directed Reading : Boosting algorithms,” 2009.
[55] W.Hu, S.Member, W.Hu, S.Maybank, and S.Member, “AdaBoost-Based Algorithm for Network,” vol. 38, no. 2, pp. 577–583, 2008.
[56] C.Seiffert, T. M.Khoshgoftaar, J.VanHulse, and A.Napolitano, “RUSBoost : A Hybrid Approach to Alleviating Class Imbalance,” IEEE Trans. Syst. Man, Cybern. - Part A Syst. Humans, vol. 40, no. 1, pp. 185–197, 2010.
[57] “Python Model Tuning Methods Using Cross Validation and Grid Search.” [Online]. Available: https://medium.com/@sebastiannorena/some-model-tuning-methods-bfef3e6544f0. [Accessed: 12-Jun-2019].
[58] B.-T.Kim et al., “Stage T1 Non–Small Cell Lung Cancer: Preoperative Mediastinal Nodal Staging with Integrated FDG PET/CT—A Prospective Study,” Radiology, vol. 241, no. 2, pp. 501–509, Nov.2006.
[59] “Basic Ensemble Learning (Random Forest, AdaBoost, Gradient Boosting)- Step by Step Explained.” [Online]. Available: https://towardsdatascience.com/basic-ensemble-learning-random-forest-adaboost-gradient-boosting-step-by-step-explained-95d49d1e2725. [Accessed: 12-Jun-2019].
[60] D.Kim, W. H.Kim, and C. G.Kim, “Dual-time-point FDG PET / CT : Is It Useful for Lymph Node Staging in Patients with Non-Small-Cell Lung Cancer ?,” pp. 196–200, 2012.
[61] T.Kasai et al., “Dual-time point scanning of integrated FDG PET/CT for the evaluation of mediastinal and hilar lymph nodes in non-small cell lung cancer diagnosed as operable by contrast-enhanced CT,” Eur. J. Radiol., vol. 75, no. 2, pp. 143–146, 2010.
[62] F.Orlhac, J.Maisonobe, C. A.Garcia, and B.Vanderlinden, “Tumor Texture Analysis in 18 F-FDG PET: Relationships Between Texture Parameters, Histogram Indices, Standardized Uptake Values, Metabolic Volumes, and Total Lesion Glycolysis,” vol. 55, no. 3, pp. 414–422, 2014.
[63] Seyhan Karacavus Bulent Yilmaz , “ Automatic Staging and Subtype Determination for Non-Small Cell Lung Carcinoma Using PET Image Texture Analysis” pp. 123–126, 2016.
[64] S.Bouzidi, F.Baldacci, and P.Desbarats, “3D segmentation of the tracheobronchial tree using multiscale morphology enhancement filter,” vol. 4617, pp. 207–214, 2016.

校內：2023-06-01公開
校外：不公開電子論文尚未授權公開，紙本請查館藏目錄

簡易檢索 / 詳目顯示

相關論文