近年來隨著醫學量化的發展，影像的分析技術逐漸被運用於臨床醫學的研究上，例如磁振造影(Magnetic Resonance Imaging，MRI)[1]為目前廣泛應用於醫學量化影像的技術之一。若能將其採集的臨床醫學影像，與特定的分類演算法結合來分析其結果，必能協助醫生判別腦部物質的改變與病灶之間的存在關係，進一步預測病人之病況[2]。
本篇論文的目的為發展一個能有效偵測白質高訊號(White Matter Hyperintensities，WMHs)的演算法。WMHs為腦醫學[3]上圍發展潛在疾病的重要依據，因此能順利偵測出該區域對於醫生診斷有莫大的幫助。演算法設計方面，我們將MRI影像視為多光譜影像(Multispectral Imaging)來分析處理。首先我們藉由波段擴張過程(Band Expansion Process，BEP)提升的影像資訊，再利用自動化目標物產生過程(Automatic Target Generation Process，ATGP)來偵測WMHs的初始位置。並設計了疊代式限制能量最小化法(Iterative Constrained Energy Minimization，ICEM)來偵測整張影像WMHs的分佈區域，同時壓抑周圍非病灶之組織。
本論文旨在將實驗結果量化。我們使用設定閾值(Threshold)將ICEM偵測結果轉為二值化影像結果，在使用相似性指標(Similarity Index，SI)來量化對病灶檢測準確率。透過使用臨床以及合成腦MRI影像的實驗可證實，我們的演算法能有效地捕捉到白質高訊號區域。此外我們也透過循序式的分析，發現所使用的BEP波段影像中若含有T1-weighted所增生的影像，就能夠有效的提升SI值。

With the rapid development of medical quantification, image analysis techniques were gradually applied to the research of clinical medicine. For example, Magnetic Resonance Imaging (MRI) is one of the image technologies that is widely used for medical imaging. If we can use appropriate image processing algorithms to analyze the MRI images, the produced results will be more objective than conventional visual evaluation. In addition, they can assist doctors to discriminate the relationship between the changes of brain’s substances and the lesions so as to predict patients’ condition.
This thesis aims to develop an algorithm that can effectively detect the White Matter Hyperintensities (WMHs) in brain MRI image. WMHs is an important basis that may cause potential diseases on the brain’s upper half in brain medicine. Therefore, if we can detect its area, it must be great aid when doctors make diagnosis. In algorithm design, we treat MRI as a kind of multispectral imaging. First, we use Band Expansion Process (BEP) to increase the amount of band images, and use Automatic Target Generation Process (ATGP) to detect the initial location of WMHs. Second, we propose an algorithm, called Iterative Constrained Energy Minimization (ICEM), which can detect the WMHs’ distribution and suppress the surrounding non-lesion organizations simultaneously.
Except for the visual results, we also design a quantification method for the experiment. We set different thresholds to convert the ICEM’s maps to binary images, then use Similarity Index (SI) to evaluate the accuracy of lesions’ examination. By virtue of the experiments conducted on both clinical and synthetic MRI image data, our algorithm can effectively find the WMHs’ distribution. Besides, through the extra experiment on progressive analysis, we find that using the BEP bands produced by T1-weighted image can significantly increase the performance of WMH’s detection.

論文審定書 i
誌謝 ii
摘要 iii
ABSTRACT iv
目錄 vi
圖目錄 ix
表目錄 xii
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機 1
1.3磁振造影 2
1.4 論文架構 6
第二章 研究方法 7
2.1平滑處理(SMOOTHING PROCESS) 7
2.2 波段擴張過程(BEP) 8
2.3限制能量最小化法(CEM) 10
2.4自動化目標物產生過程(ATGP) 11
2.5 序列式波段選擇法(SBS) 12
2.6 循序式波段選擇法(PBS) 14
2.7 疊代式限制能量最小化法(ICEM) 16
2.8圖形使用者介面(GUI) 17
2.9大津演算法(OTSU’S) 18
第三章 實驗方法與結果 19
3.1實際腦部臨床影像(REAL BRAIN IMAGE) 19
3.1.1影像介紹 19
3.1.2病灶分析：使用ATGP偵測病灶之位置與ICEM偵測結果 21
3.1.3病灶分析：使用平滑處理改善ATGP與ICEM結果 25
3.1.4病灶分析：不同版本之MATLAB結果比較 32
3.1.5病灶分析：將ICEM結果做絕對值之比較 42
3.1.6閾值計算：使用圖形使用者介面選擇閾值 46
3.1.7總結 50
3.2 合成腦部影像(SYNTHETIC BRAIN IMAGE) 51
3.2.1影像介紹 51
3.2.2病灶分析：使用ATGP偵測病灶之位置 53
3.2.3病灶分析：使用平滑處理與ATGP偵測病灶點位置 54
3.2.4病灶分析：病灶偵測演算法於不同閾值之量化比較 55
3.2.5使用循序式與序列式波段選擇法分析分析各種BEP組合的相似性 59
3.2.6實驗修正：使用完整三種權重之BEP波段進行病灶分析 72
3.2.7參照循序式與序列式波段選擇法之結果做ICEM 79
3.2.8總結 88
第四章 結論與未來展望 89
4.1 結論 89
4.2 未來展望 90
參考文獻 91
附錄A-使用ATGP偵測病灶位置與ICEM偵測結果(REAL BRAIN IMAGE) 93
附錄B-使用平滑處理改善ATGP與ICEM結果(REAL BRAIN IMAGE) 96
附錄C-使用ATGP偵測病灶位置(SYNTHETIC BRAIN IMAGE) 97
附錄D-使用平滑處理改善ATGP偵測病灶之位置(SYNTHETIC BRAIN IMAGE) 98
附錄E-病灶偵測演算法於不同閾值之量化比較 99
附錄F-使用循序式與序列式波段選擇法分析分析各種BEP組合的相似性 103
附錄G-使用完整三種權重之BEP波段進行病灶分析 117

[1] C. Westbrook, "MRI in practice / Catherine Westbrook and Caroline Kaut Roth with John Talbot," ed: Malden, MA : Blackwell Pub, 2005.
[2] H. Jokinen, N. Goncalves, R. Vigario, J. Lipsanen, F. Fazekas, R. Schmidt, et al., "Early-Stage White Matter Lesions Detected by Multispectral MRI Segmentation Predict Progressive Cognitive Decline," Front Neurosci, vol. 9, p. 455, 2015.
[3] 吳維, 解剖學: 台北市 : 合記, 1987.
[4] T. Samaille, L. Fillon, R. Cuingnet, E. Jouvent, H. Chabriat, D. Dormont, et al., "Contrast-based fully automatic segmentation of white matter hyperintensities: method and validation," PLoS One, vol. 7, p. e48953, 2012.
[5] C. C.-C. C. e. al, "Intra-pixel multispectral processing of magnetic," Int. J. Computational Science and Engineering, vol. 8, 2013.
[6] J.-Y. Huang, "Brain Lesions Detection Using Constrained Energy Minimization for MRI," M.E. dissertation,National Chung Hsing University, 2013.
[7] R. R. Edelman and S. Warach "Magnetic Resonance Imaging," New England Journal of Medicine, vol. 328, pp. 708-716, 1993.
[8] 盧家鋒. (2013). 磁振影像基本原理：組織對比. Available: http://www.ym.edu.tw/~cflu/CFLu_course_mriprinp.html
[9] R. E. W. Rafael C. Gonzalez, Steven L. Eddins,譯者: 王振仲 黃健興, Digital Imaging Processing Using MATLAB.數位影像處理使用MATLAB, 2012.
[10] R. E. W. 著. 繆. 譯. Rafael C. Gonzalez, Digital Imaging Processing 數位影像數理, 2009.
[11] X. Jiao and C.-I. Chang, "Kernel-based constrained energy minimization (K-CEM)," 2008, pp. 69661S-69661S-11.
[12] C. -I. Chang, C. Gao, and S. Y. Chen, "Recursive Automatic Target Generation Process in Subpixel Detection," IEEE Geoscience and Remote Sensing Letters, vol. 12, pp. 1848-1852, 2015.
[13] C.-I. Chang, Hyperspectral Imaging: Techniques for Spectral Detection and Classification, 1 ed.: Springer US, 2003.
[14] C.-H. L. Y.-W. Wang, K.-H. Liu, "Sequential Band Selection Method Based on Band Vector De-Correlation for Progressive Data Analysis of Hyperspectral Imagery," presented at the CVGIP, 2015.
[15] C. -I. Chang, C. C. Wu, W. Liu, and Y. C. Ouyang, "A New Growing Method for Simplex-Based Endmember Extraction Algorithm," IEEE Transactions on Geoscience and Remote Sensing, vol. 44, pp. 2804-2819, 2006.
[16] C.-I. Chang, "Hyperspectral Data Processing: Algorithm Design and Analysis, 1 edition," ed: Plenum Publishers, 2003.
[17] Q. Du and H. Yang, "Similarity-Based Unsupervised Band Selection for Hyperspectral Image Analysis," IEEE Geoscience and Remote Sensing Letters, vol. 5, pp. 564-568, 2008.
[18] J. A. Tropp and A. C. Gilbert, "Signal Recovery From Random Measurements Via Orthogonal Matching Pursuit," IEEE Transactions on Information Theory, vol. 53, pp. 4655-4666, 2007.
[19] C. -I. Chang and K. H. Liu, "Progressive Band Selection of Spectral Unmixing for Hyperspectral Imagery," IEEE Transactions on Geoscience and Remote Sensing, vol. 52, pp. 2002-2017, 2014.
[20] C.-I. Chang, Hyperspectral data processing: algorithm design and analysis: John Wiley & Sons, 2013.
[21] wikipedia. Skewness. Available: https://en.wikipedia.org/wiki/Skewness
[22] H. M. Chen, C. Lin, S. Y. Chen, C. H. Wen, C. C. C. Chen, Y. C. Ouyang, et al., "PPI-SVM-Iterative FLDA Approach to Unsupervised Multispectral Image Classification," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 6, pp. 1834-1842, 2013.
[23] N. Otus, "A Tlreshold Selection Method from Gray-Level Histograms," IEEE Transactions on System ,Man , and Cybernetics, vol. 9, pp. 62-66, 1979.
[24] T. J. Grabowski, R. J. Frank, N. R. Szumski, C. K. Brown, and H. Damasio, "Validation of Partial Tissue Segmentation of Single-Channel Magnetic Resonance Images of the Brain," NeuroImage, vol. 12, pp. 640-656, 12// 2000.
[25] MRIcro. Available: http://people.cas.sc.edu/rorden/mricro/index.html
[26] BrainWeb Data. Available: http://brainweb.bic.mni.mcgill.ca/brainweb/
[27] 李三剛. (2009) 應用“獨立成份分析法”改善腦部MR 影像組織分割績效之研究. 中華放射醫誌 Chin J Radiol 93-101.
[28] 傅家啓. (2005) 脈衝耦合類神經網路於磁振影像之腦組織分割. 中華放射醫誌 Chin J Radiol.
[29] wikipedia. Similarity Index. Available: https://en.wikipedia.org/wiki/Jaccard_index