臺灣博碩士論文加值系統

肺部病變是肺癌的初期徵兆。根據2016年癌症統計報告，近年來肺癌罹患率與死亡率不斷攀升，晚期的5年存活率僅19%。因此經由X光影像或電腦斷層影像發現早期病變區域，進行追蹤檢查，成為降低死亡風險的方法之一。
本論文提出利用深度學習，預測胸腔電腦斷層影像中可能的病變區域，並將區域框出，作為醫護人員診斷報告中的參考資訊，以期能夠及早發現病變區域。深度學習模型主要基於Mask R-CNN，採用雙損失函數，分別對預測位置的中心距離和長寬大小進行修正。實驗結果顯示，相較於僅使用單一損失函數對兩項評斷標準，分別使用兩種不同損失函數，有更好的判讀效果。

Lung lesion is the symptom of lung cancer. In the report of ‘CANCER REGISTRY ANNUAL REPORT, 2016’ in Taiwan, the morbidity and mortality rate of lung cancer are constantly increasing in recent years. The five-year survival rate of patient in stage III or IV has decreased to 19%. Using X -ray or computer tomography to discover the lesion regions in early stage and trace exam is an effect method to decrease the mortality rate.
This thesis utilizes deep learning to predict the lesion regions that carries high risks in chest computer tomography and mask the lesion regions. Healthcare professional can take the results of prediction as reference in diagonal report to discover lung lesion in early stage. The deep learning model is based on Mask R-CNN with twin loss functions to calculate the distance of center and the difference of width and height, respectively. In the results of experience, it gets better performance with using two different loss functions than only one loss function.

摘 要.......................................i
Abstract....................................ii
Acknowledgements............................iii
Table of Content............................iv
List of Tables..............................vi
List of Figures.............................vii
List of Abbreviation........................xi
Chapter1 Introduction.......................1
1.1 Overview................................1
1.2 Organization of the Thesis..............4
Chapter2 Background and Related Works.......5
2.1 Computer Tomography (CT)................5
2.1.1 Projection and Radon Transform........7
2.1.2 Image Reconstruction..................11
2.2 Windowing...............................16
2.2.1 Window Level..........................17
2.2.2 Window Width..........................18
2.3 Convolutional Neural Network............19
2.3.1 Back-propagation ＆ Gradient Descent...21
2.3.2 Learning rate.........................22
2.4 Loss Function...........................23
2.4.1 L1 Loss...............................23
2.4.2 L2 Loss...............................24
2.4.3 Huber Loss............................26
2.4.4 Cauchy’s Loss Function................27
2.5 Transfer learning.......................28
2.6 Mask R-CNN..............................29
2.6.1 Feature Pyramid Network (FPN).........31
2.6.2 Region Proposal Network (RPN).........32
2.6.3 RoIPool and RoIAlign..................34
Chapter3 The Proposed Algorithm.............36
3.1 Intensity Preprocessing.................38
3.2 Network Training........................38
3.2.1 Model Weights Pre-training............39
3.2.2 Main Model Training...................41
3.3 Loss Function...........................41
3.4 Test Method.............................46
Chapter4 Results and Discussion.............48
4.1 Experimental Dataset and Setting........48
4.2 Performance Evaluation..................50
Chapter5 Conclusion and Future Works........59
5.1 Conclusion..............................59
5.2 Future Works............................59
Reference...................................61

[1]Health Promotion Administration Ministry of Health and Welfare, “CANCER REGISTRY ANNUAL REPORT, 2016, December 2018.
[2]Callister ME, Baldwin DR, Akram AR, et al. British Thoracic Society guidelines for the investigation and management of pulmonary nodules., Thorax 2015; 70: Suppl. ii1–ii54.
[3]American Cancer Society, Cancer Facts ＆ Figures 2019, pp18, 2019.
[4]DeSantis CE, Miller KD, Goding Sauer A, Jemal A, Siegel RL. Cancer statistics for African Americans, 2019, 14 February 2019.
[5]Pao-Shan Wang, and Kuo-Chen Cheng, Management of Pulmonary Nodules Detected on CT Image,J Intern Med Taiwan(JIMT), Jan,2018: pp.137-142.
[6]James C. Reed , “Chest Radiology: Plain Film Patterns and Differential Diagnoses,ISBN:978-1-4377-2345-8, December 2017.
[7]Kirchner, “Chest Radiology: A Resident's Manual,ISBN:978-3-1315-3871-0, March 2011.
[8]Jannette Collins, Eric J. Stern , “Chest Radiology: The Essentials, ISBN:978-1-4511-4448-2, 2014.
[9]Lynn T. Tanoue, MD and Frank C Detterbeck, MD, FACS, FCCP “Lung Cancer: A Practical Approach to Evidence-Based Clinical Evaluation and Management, ISBN: 978-0-323-48565-4, 2018.
[10]Clarus Leungcorresponding, Tawimas Shaipanich, “Current Practice in the Management of Pulmonary Nodules Detected on Computed Tomography Chest Scans, Canadian Respiratory Journal,6 Jan 2019.
[11]A. C. Kak and Malcolm Slaney, “Principles of Computerized Tomographic Imaging, Society of Industrial and Applied Mathematics, 2001.
[12]Lim W, Ridge CA, Nicholson AG, Mirsadraee S. “The 8th lung cancer TNM classification and clinical staging system: review of the changes and clinical implications, Quant Imaging Med Surg. 2018, pp. 709–718.
[13]Steven E Weinberger. “Diagnostic evaluation and management of the solitary pulmonary nodule., 2017 March.
[14]Wim van Drongelen Signal Processing for Neuroscientists - Chapter 7 - 1-D and 2-D Fourier Transform Applications., Second Edition, pp. 119-152, 2018.
[15]Kaiming He, Georgia Gkioxari, Piotr Dollar, Ross GirshickMask “Mask R-CNN, The IEEE International Conference on Computer Vision, pp. 2961-2969, 2017.
[16]Ross Girshick, Fast R-CNN, 2015 Sep.
[17]Ke Yan, Xiaosong Wang, Le Lu, Ronald M. Summers, DeepLesion: Automated Mining of Large-Scale Lesion Annotations and Universal Lesion Detection with Deep Learning, Journal of Medical Imaging 5(3), pp. 036501, 2018.
[18]Mark Everingham, Luc Van Gool, Christopher K. I. Williams, John Winn, Andrew Zisserman, The Pascal Visual Object Classes (VOC) Challenge, International Journal of Computer Vision Volume 88 Issue 2, pp. 303–338, June 2010.
[19]V. Nair and G E Hinton “Rectified linear units improve restricted boltzmann machines in Processing of the 27th International Conference on International Conference on Machine Learning Haifa, Israel, 2010, pp. 807-814.
[20]Xuelong Li, Quanmao Lu, Yongsheng Dong, Dacheng Tao, “Reference: Robust Subspace Clustering by Cauchy Loss, IEEE Transactions on Neural Networks and Learning Systems, 2018.
[21]Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, Serge Belongie “Feature Pyramid Networks for Object Detection, The IEEE CCVPR, 2017, pp. 2117-2125.
[22]Zhao, Tingting, Wang, Huafeng, Li, Lihong, Qi, Yifang, Haoqi, Gao, Han, Fangfang, Liang, Zhengrong, Qi, Yanmin, Cao, Yuan, “A Hybrid CNN Feature Model for Pulmonary Nodule Differentiation Task, BIVPCS 2017 and POCUS 2017, pp. 19-2614, Sep. 2017.
[23]Hoo-Chang Shin, Holger R. Roth, Mingchen Gao, Le Lu, Ziyue Xu, Isabella Nogues, Jianhua Yao, Daniel Mollura, and Ronald M. Summers, “Deep Convolutional Neural Networks for Computer-Aided Detection: CNN rchitectures, Dataset Characteristics and Transfer Learning, IEEE Transactions on Medical Imaging, Vol. 35, No. 5, May 2016.
[24]Feuerstein, Marco, Deguchi, Daisuke, Kitasaka, Takayuki, Iwano, Shingo, Imaizumi, Kazuyoshi, Hasegawa, Yoshinori, Suenaga, Yasuhito, Mori, Kensaku. “Automatic mediastinal lymph node detection in chest CT., Proceedings of SPIE - The International Society for Optical Engineering, Feb, 2009.
[25]Masakazu Urata, Yuko Kijima, Munetsugu Hirata, Yoshiaki Shinden, Hideo Arima, Akihiro Nakajo, Chihaya Koriyama, Takaaki Arigami, Yoshikazu Uenosono, Hiroshi Okumura, Kosei Maemura, Sumiya Ishigami, Heiji Yoshinaka and Shoji Natsugoe, “Computed tomography Hounsfield units can predict breast cancer metastasis to axillary lymph nodes, Urata et al. BMC Cancer 2014.
[26]Holger R. Roth, Le Lu, Jiamin Liu, Jianhua Yao, Ari Seff, Kevin Cherry, Lauren Kim, and Ronald M. Summers, “Improving Computer-Aided Detection Using Convolutional Neural Networks and Random View Aggregation., IEEE Transactions on Medical Imaging, Vol. 35, No. 5, May 2016.