臺灣博碩士論文加值系統

癌症為台灣十大死亡原因第一名，肺癌又為癌症之首。現階段以LDCT(低劑量電腦斷層掃描)為篩選肺癌病灶最有效的方法。但LDCT的判讀，相當依賴醫師個人經驗。
此研究的目的為開發電腦軟體，從LDCT影像自動化篩選出結節位置。方法為將電腦斷層掃描的灰階圖檔讀入後，經由撰寫的Matlab程式篩選出候選圖案，並針對個別的候選圖案計算12項參數值，包含面積、周長、HU平均值、HU值標準差、Shape(周長的平方除以面積，判別外觀圓滑度)、Entropy、Spiculation(毛針刺為惡性結節之特徵)、Peaks(HU值之 local maximum 個數)、C.V.變異係數、標準差除以面積、Calcify鈣化指數(HU值大於一定值視為鈣化)、First_deri(各pixel HU值之一階微分除以面積，判別其凌亂程度)。計算後之參數，導入人工智慧類神經網路，判別其為結節或非結節。最後將其結果比對醫生所提供的病例資訊與位置。我們使用36個測試候選圖案，經類神經網路判讀，成功判讀出結節4個，非結節28個，誤判結節為非結節2個，誤判非節結為結節2個，其真陽性機率(Sensitivity)為66.7%，真陰性機率(Specificity)為93.3%，準確率(Accuracy)為88.9%。本研究可協助醫生篩選判別肺部結節，有助於肺部病變的診斷與治療。

Cancer is the first of top ten death causes in Taiwan, and lung cancer is the most serious cancer. LDCT (Low Dose Computed Tomography) is an effective examination technology contemporarily. However, finding the lesions heavily relies on medical physician’s judgements.
This study develops a computer software that can automatically screens pulmonary nodules from LDCT. The program, developed with Matlab, filters the LDCT figures and calculates 12 parameters of each candidate parts. The parameters are area, perimeter, HU mean, standard deviation of HU, shape, entropy, spiculation, peak, coefficient of variation of HU, standard deviation divided by area, calcify, first derivative of HU. And then, we import the parameters into an artificial neural network to identify the candidates are nodules or non-nodules. The outcomes of the artificial neural network are compared with experienced physician judgements. Employing 36 candidate figures as the test group, the neural network successfully identifies 28 non-nodules and 4 nodules, misidentifies 2 nodules as non-nodules and 2 non-nodules as nodules. The result is 66.7% sensitivity and 93.3% specificity, and the accuracy is 88.9%. Our work can help physicians with screening the chest nodules and their diagnosis and treatments.

摘 要 i
ABSTRACT ii
致謝 iii
表目錄 vi
圖目錄 vii
第一章 序 論 1
第二章 文獻探討 2
2.1電腦斷層掃描 2
2.2類神經網路 4
2.3文獻討論 7
第三章 研究方法 9
3.1電腦斷層基本名詞介紹 9
3.2影像讀取之前置作業 11
3.3候選圖案篩選 15
3.4參數計算 16
3.5 倒傳遞類神經網路 (BPNN) 19
3.6 研究流程 20
第四章 研究結果 23
4.1 Case_1 23
4.2 Case_2 25
4.3 Case_3 26
4.4 Case_4 27
4.5 Case_5 28
4.7 Case_7 32
4.8 Case_8 33
4.9 Case_9 35
4.10 Case_10 36
4.11 Case_11 38
4.12 Case_12 39
4.13 Case_13 40
4.14 Case_14 41
4.15 Case_15 42
4.16 Case_16 43
4.17 Case_17 44
4.18 Case_18 45
第五章 結果討論 46
5.1 數據分析 46
5.2 類神經結果分析 50
5.3 討論 52
參考文獻 53

[1]低劑量CT篩檢，聯合報記者鄧桂芬、陳雨鑫台北報導，2015。
[2]中國醫藥大學醫學系解剖科&中國醫藥大學附設醫院放射部
http://www2.cmu.edu.tw/~cmcmd/ctanatomy/imagetool/ctintro.html
[3]Jinzhu Yang, Lin Lu and Shuang Ma, “ An Automatic Method to Extract Brain Tissue in CT Data, ” Information Technology in Medicine and Education (ITME), 2016 8th International Conference , Dec. 2016.
[4]CT Scan (Computerized Tomography, CAT Scan).
http://www.medicinenet.com/cat_scan/article.htm
[5]Somayeh Molaei, Frederick K. Korley and S.M. Reza Soroushmehr, “ A machine learning based approach for identifying traumatic brain injury patients for whom a head CT scan can be avoided, ” 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) , Aug. 2016.
[6]Advantages & Disadvantages of CT Scans
http://www.ehow.com/list_6456817_advantages-disadvantages-ct-scans.html
[7]張斐章、張麗秋，“類神經網路導論原理與應用第二版”，2015滄海圖書資訊股份有限公司出版 ，pp. 1-297.
[8]Stefan Diederich, Dag Wormanns, and Michael Semik, “Screening for EarlyLung Cancer with Low-Dose Spiral CT: Prevalence in 817 Asymptomatic Smokers1,” Published online before print 10.1148/radiol.2223010490, Radiology 2002; 222:773–781, Sept. 2016.
[9]Emre Dandll, Murat Cakiroglu, Ziya Eksi, Murat Ozkan, Ozlem Kar Kurt and Arzu Canan, “Artificial Neural Network-Based Classification System for Lung Nodules on Computed Tomography Scans, ” International Conference of Soft Computing and Pattern Recognition, Aug. 2014.
[10]Jhilam Mukherjee, Amlan Chakrabarti and Soharab Hossain Skaikh, “Automatic Detection and Classification of Solitary Pulmonary Nodules from Lung CT Images, ” 2014 Fourth International Conference of Emerging Applications of Information Technology, Dec. 2014.
[11]Matlab Image Processing Toolbox User's Guide.


[12]T. Arumuga Maria Devi, V. I. Mebin Jose and P. Kumar Parasuraman “ A novel approach for automatic detection of non-small cell lung carcinoma in CT images, ” Control, Instrumentation, Communication and Computational Technologies (ICCICCT), 2016 International Conference, Dec. 2016.
[13]杜書儒助理教授，“微米級電腦斷層掃描造影系統”，醫學物理暨影像科學研究所。
http://enews.cgu.edu.tw/files/16-1068-42758.php?Lang=zh-tw 。
[14]鄭高珍醫師、蔣士仁醫師(私人通訊)，台南奇美醫院胸腔外科。
[15]Wiilliam J.Palm III 著，歐陽彥杰 審閱，呂明和、黃逸群 譯，“Matlab 6 在工程上的應用”，2003年美商麥格羅希爾國際股份有限公司出版。
[16]Jhilam Mukherjee, Amlan Chakrabarti, Soharab Hossain Shaikh and Madhuchanda Kar, “Automatic Detection and Classification of Solitary Pulmonary Nodules from Lung CT Images,” 2014 Fourth International Conference of Emerging Applications of Information Technology, pp. 294-299, Dec. 2014.
[17]胡維平，“熱力學性質計算”，國立中正大學化學暨生物化學系。
[18]Maximum Entropy Modeling http://homepages.inf.ed.ac.uk/lzhang10/maxent.html
[19]施穎銘、林聖皓，“低劑量電腦斷層掃描用於肺癌的篩檢”，彰化基督教醫院胸腔內科。
[20]FDG PET/CT 正子電腦斷層在癌症的應用
https://sites.google.com/site/cboatpetct/home/fulltext/chapter-07/chapter-07-02
[21]Khaing Zin Htwe, Kunihito YAMAMORI, Tetsuro KATAYAMA and Tin Mar Kyi, ’’ Automated Lung Nodule Classification by Artificial Neural Network and Fuzzy Inference System,’’ 2016 IEEE 5th Global Conference on Consumer Electronics, Oct. 2016.
[22]Mustafa Alam, Ganesh Sankaranarayanan and Venkat Devarajan, ‘’Lung Nodule Detection and Segmentation Using a Patch-Based Multi-Atlas Method,’’ 2016 International Conference on Computational Science and Computational Intelligence, Dec. 2016.
[23]實證相關名詞，馬偕醫院醫學教育部
http://www.mmh.org.tw/taitam/medical_edu/www/?contentID=644