磁振造影是一種常用的醫學影像之診斷模式,它對軟組織有較佳的顯像效果。本論文使用的磁振造影有：藥物動力學磁振造影及功能性磁振造影。藥物動力學磁振造影是在打藥之後,在連續時間取得一系列影像,它藉由在不同時間點所取得的影像之訊號強度變化來診斷疾病。功能性磁振造影則產生腦神經激活之影像,它可用來研究腦部之結構、功能及病理的關係。由於患者在造影時移動,連續取得的影像並沒有對準。因此在本論文,校準連續取得的磁振造影影像是項基礎之課題。最小平方估計法是一標準的參數估計之方法。無論如何,異常點可能存在(由於non-Gaussain雜訊、生理組織病變、與移動相關之製品、…等等),並且因此使得移動參數估計之結果有較大的計算誤差。本論文描述二個強韌的估計演算法,它們被用來校準連續取得的的磁振造影影像。第一個估計演算法是基於Newton方法並且使用Tukey's biweight 目標函數。第二個估計演算法是基於Levenberg-Marquardt技巧並且使用skipped mean 目標函數。這個強韌的M-估計式子藉由權數函數,給予異常點非常小的權值或者沒有權值來隱藏或拒絕異常點對系統之影響。本論文應用強韌的估計技巧至藥物動力學磁振造影及功能性磁振造影之影像校準。實驗結果顯示,強韌的估計之影像校準誤差在一個像素以內。磁振造影已被應用至鼻竇乳頭狀瘤之診斷。無論如何,由於手術後的復發或良性變化,使得手術後之影像變得更為複雜。因此,傳統的診斷方式可能會有將復發性腫瘤誤認為是良性變化,或者將良性變化誤認為是復發性腫瘤誤之情況發生。近年來,藥物動力學磁振造影已被成功地運用於直腸癌及子宮頸癌之復發性或良性變化的辨識。而鼻咽部位之腫瘤是亞洲人常患的疾病。因此,本論文評估藥物動力學磁振造影於鼻咽癌及復發性鼻竇乳頭狀瘤之診斷的可行性。 本論文利用藥物動力學磁振造影影像及強韌的移動校準來分辨復發性鼻竇乳頭狀瘤及手術後良性變化。首先,強韌的估計式子被併入非線性的最小化方法,來校準藥物動力學磁振造影影像。接著,使用者概略的選取有興趣的區域,並且使用活動的輪廓之技巧,來抽取更精確的有興趣的區域。然後,使用相對訊號增加模式,來對藥物動力學磁振造影影像的資料做相對訊號增加之計算。三參數數學模式被用來來進行藥物動力分析,並且使用非線性的最小平方適合法,來計算這些藥物參數: A(增加振幅) 及Tc(生理組織分佈時間)。這些經過計算的藥物參數A及Tc,被用來對生理組織賦予特徵。藥物動力分析顯示復發性鼻竇乳頭狀瘤比良性變化有較快的生理組織分佈時間(41相對於88),及較高的增加振幅(2.4相對於1.2)。在分辨鼻竇乳頭狀瘤及良性變化時,以生理組織分佈時間設定為65秒並且增加振幅設定為1.6當成門檻時,可以產生100% 之辨識精確度。 無論如何,上述之方法屬於半自動的切割,需要有經驗之放射線醫師概略的選取有興趣的區域。為了診斷鼻咽癌及改進上述缺點,本論文提出自動化標示鼻咽癌之方法。這個方法是由多個處理階段組成,包含:強韌的移動校準、對藥物動力學磁振造影影像的資料做相對訊號增加之計算、概略的切割、及對概略的切割精細至最後的結果。有三個方法被用來對藥物動力學磁振造影影像的資料做相對訊號增加之計算,它們是：相對訊號增加方法、斜率方法和相對訊號改變方法。在對藥物動力學磁振造影影像的資料做訊號增加之計算以後,概略的鼻咽癌之輪廓是被決定,然後使用形態學運算來對概略的切割精細至最後的結果。實驗結果顯示,本論文提出的方法所發現之鼻咽癌區域與放射線醫師手繪之鼻咽癌區域比對的結果有85% 精確度。無論如何,所提出之方法能夠有效的而且較快的確認鼻咽癌之區域。 本論文提出強韌移動校準之方法、電腦化鼻咽癌診斷之方法及區分復發性鼻竇乳頭狀瘤及手術後良性變化之方法。實驗結果證明,所提出的方法能夠提供臨床醫師一個快速且有價值之診斷工具。

Magnetic resonance imaging (MRI) is one of medical images used by doctors for diagnosing diseases. MRI shows higher quality in displaying soft tissues and tumors. Pharmacokinetic dynamic gadolinium-enhanced MR imaging and functional MR imaging (fMRI) were used in this dissertation. Dynamic MR images are obtained using fast spin-echo sequences at consecutive time after the injection of gadolinium-diethylene-triamine penta-acetic (Gd-DTPA) acid. A pharmacokinetic model analyzes time-signal intensity curves of suspected lesions. Functional MR imaging produces images of activated brain regions by detecting the indirect effects of neuronal activity on local blood volume, flow, and oxygen saturation. Thus it is a promising tool for further understanding the relationships between brain structure, function, and pathology. Because of patients' movement during imaging, serially acquired MR images do not correspond in the same pixel position. Therefore, matching corresponding points from MR images is one of fundamental tasks in this dissertation. Least-squares estimation is a standard method for parameter estimation. However, outliers (due to non-Gaussian noise, lesion evolution, motion-related artifacts, etc.) may exist and thus may cause the motion parameter estimation result to deteriorate. In this dissertation, we describe two robust estimation algorithms for the registration of serially acquired MR images. The first estimation algorithm is based on the Newton method and uses the Tukey's biweight objective function. The second estimation algorithm is based on the Levenberg-Marquardt technique and uses a skipped mean objective function. The robust M-estimators can suppress the effects of the outliers by scaling down their error magnitudes or completely rejecting outliers using a weighting function. Experimental results show the accuracy of the proposed robust estimation algorithms is within subpixel.
MR imaging has been used to evaluate nasal papilloma. However, postoperative MR imaging of nasal papilloma becomes more complicated because repair with granulation and fibrosis occurs after surgery. Therefore, it is possible to misclassify recurrences as postoperative changes or to misclassify postoperative changes as recurrences. Recently, dynamic gadolinium-enhanced MR imaging with pharmacokinetic analysis has been successfully used to identify the post-treatment recurrence or postoperative changes in rectal and cervical carcinoma. Nasopharyngeal carcinoma (NPC) comprising malignant tumors is a disease more common in Asia than in other parts of the world. Hence, in this dissertation, we evaluate the feasibility of dynamic gadolinium-enhanced MR imaging with pharmacokinetic analysis in detecting NPC and distinguishing recurrent nasal papilloma from postoperative changes (fibrosis or granulation tissue).
In this dissertation, a new approach to differentiate recurrent nasal papilloma from postoperative changes using pharmacokinetic dynamic gadolinium-enhanced MR imaging and robust motion correction is presented. First, a robust estimation technique is incorporated into nonlinear minimization method to robustly register dynamic gadolinium-enhanced MR images. Next, user roughly selects the region of interest (ROI) and an active contour technique is used to extract a more precise ROI. Then, the relative signal increase (RSI) is calculated. We use a three-parameter mathematical model for pharmacokinetic analysis. The pharmacokinetic parameters A (enhancement amplitude) and Tc (tissue distribution time) are calculated by a nonlinear least-squares fitting technique. The calculated A and Tc are used to characterize tissue. Pharmacokinetic analysis shows that recurrent nasal papilloma has faster tissue distribution time (Tc, 41 versus 88 seconds) and higher enhancement amplitude (A, 2.4 versus 1.2 arbitrary units) than do postoperative changes. A cut-off of 65 seconds for tissue distribution time and 1.6 units for enhancement amplitude yields an accuracy of 100% for differentiating recurrent nasal papilloma from postoperative changes.
Though the above methods obtained good results, finding the region of interest (ROI) was done in a semi-automatic manner. For diagnosing NPC and improve the drawback, a system that automatically detects and labels NPC with dynamic gadolinium-enhanced MR imaging is presented. This system is a multistage process, involving motion correction, gadolinium-enhanced MR data quantitative evaluation, rough segmentation, and rough segmentation refinement. Three approaches, a relative signal increase method, a slope method and a relative signal change method, are proposed for the quantitative evaluation of gadolinium-enhanced MR data. After the quantitative evaluation, a rough NPC outline is determined. Morphological operations are applied to refine the rough segmentation into a final mask. The NPC detection results obtained using the proposed methods had a rating of 85% in match percent compared with these lesions identified by an experienced radiologist. However, the proposed methods can identify the NPC regions quickly and effectively.
In this dissertation, the proposed methods provide significant improvement in correcting the motion-related artifacts and can enhance the detection of real brain activation and provide a fast, valuable diagnostic tool for detecting NPC and differentiating recurrent nasal papilloma from postoperative changes.

中文摘要………………….…………………………………………………………….…i-1
Abstract…………………………………………..…….…………………………………ii-1
Acknowledgements…………………………..…….…………………….……….……iii-1
Table of Contents………...…………………………………………………………….iv-1
List of Figures…………………..…………………………………………………….….v-1
List of Tables…………………………...……………………………….……………….vi-1
1 Introduction…………………………………………………………….………….1-1
1.1 Motivation and Recent Related Works………………………..…………..1-1
1.2 Summary of the Dissertation………………………………...…….………….1-6
1.3 Organization of the Dissertation………………………………….………….1-8
2 Mathematical Preliminaries……………………………………………………2-1
2.1 Least-squares Estimation for Nonlinear Problem…………….………...2-1
2.2 Parameter Estimation based on the Newton Method…….……….………2-2
2.2.1 Traditional Newton Method………………………………….………2-2
2.2.2 Robust Estimation for the Newton Method……………....……2-3
2.3 Parameter Estimation based on the Levenberg-Marquardt Method…...2-6
2.3.1 Traditional Levenberg-Marquardt Method…………….…………..2-6
2.3.2 Robust Estimation for the Levenberg-Marquardt Method…….…..2-7
3 Applications of Robust Motion Correction.……..……………………..…….3-1
3.1 Methods……………..………………………..……………………….………….3-1
3.2 Dynamic Gadolinium-Enhanced MR Imaging Application………...…3-3
3.2.1 Experimental Results…………………..…………………..…………….3-3
3.3 Functional MR Imaging of the Brain Application………..….….…..……3-7
3.3.1 Experimental Results……………………………………..…………….3-7
3.3.2 Discussion…………………………………..…………………………...3-15
4 Distinguishing Recurrent Nasal Papilloma from Postoperative Changes with Dynamic Gadolinium-Enhanced MR Imaging………………...……..4-1
4.1 Patients and Imaging Schemes…………………..………………………..4-1
4.2 Preliminaries………………………………………………………………..…4-2
4.3 System Overview………………………………………………………..…4-3
4.4 Boundary Detection of the Region of Interest…………………………..4-4
4.5 Quantitative Evaluation of Dynamic MR Data…….……………………4-5
4.5.1 Relative Signal Increase Method…………………………….………4-5
4.5.2 Color Representation of the Region of Interest….………………..4-6
4.6 Experimental Results…………………………………………….……………4-7
4.7 Discussion……………………..……………………………………………….4-13
5 Automated Nasopharyngeal Carcinoma Detection with Dynamic Gadolinium-Enhanced MR Imaging…………………………….…………5-1
5.1 Patients and Imaging Scheme…………………………………..………..5-1
5.2 System Overview……………………………………….………………….5-2
5.3 Generate an Initial Head Mask………………….……………………….5-3
5.4 Quantitative Evaluation of Dynamic MR Data…………………….……5-5
5.4.1 Relative Signal Increase Method………………………………………5-5
5.4.2 Slope Method…………………………………………..……………….5-7
5.4.3 Relative Signal Change Method……………………………….…….5-8
5.5 Rough Segmentation………………………………………………………5-10
5.6 Refinement of the Rough Segmentation………………………………...5-13
5.7 Experimental Results…………………………………………….…………5-18
5.8 Discussion……..……………………………………………….…………….5-23
6 Conclusions and Future Works……………………………………………...6-1
References………………………………………………………………….…………..vii-1

[Alexander, 1996] Alexander ME and Somorjai RL: The registration of MR images using multiscale robust methods. Magnetic Resonance Imaging, 14(5): 453-468, 1996.
[Alexander, 1997] Alexander ME, Scarth G, and Somorjai RL: An improved robust hierarchical registration algorithm. Magnetic Resonance Imaging, 15(4): 505-514, 1997.
[Alpert, 1996] Alpert NM, Berdichevsky D, Levin Z, Morris ED, and Fishman AJ: Improved methods for image registration. Neuroimage, 3: 10-18, 1996.
[Barra, 2001] Barra V and Jean-Yves B: Segmentation of muscle and fat comparements from MR images using a fuzzy clustering algorithm. Proceedings of ISMRM Joint Annual Meeting, Scotland, April 21-27, 849, 2001.
[Black, 1996] Black MJ and Rangarajan A: On the unification of line processes, outliers rejection and robust statistics in early vision. International Journal of Computer Vision, 19(1): 57-91, 1996.
[Brix, 1991] Brix G, Semmler W, Port R, Schad LR, Layer G, and Lorenz WJ: Pharmacokinetic parameters in CNS Gd-DTPA enhanced MR imaging. Journal of Computer Assisted Tomography, 15: 621-628, 1991.
[Brown, 1992] Brown LG: A survey of image registration techniques. ACM Computing Surveys, 24(4): 325-376, 1992.
[Chakraborty, 1996] Chakraborty A, Staib LH, and Duncan JS: Deformable boundary finding in medical images by integrating gradient and region information. IEEE Trans. Med. Imag., 15(6): 859-870, 1996.
[Chalana, 1996] Chalana V, Linker T, Haynor DR, and Kim Y: A multiple active contour model for cardiac boundary detection on echocardiographic sequences. IEEE Trans. Med. Imag., 15(3): 290-298, 1996.
[Clark, 1998] Clark MC, Hall LO, Goldgof DB, Velthuizen R, Murtagh FR, and Silbiger MS: Automatic tumor segmentation using knowledge-based techniques. IEEE Trans. Med. Imag., 17(2): 187-201, 1998.
[Coiras, 2000] Coiras E, Santamaria J, and Miravet C: Segment-based registration technique for visual-infrared images. Optical Engineering, 39: 282-289, 2000.
[Cox, 1996] Cox RW: Motion and functional MRI. Informal notes for the Boston'96 Workshop on Functional MRI, 1996.
[Davatzikos, 1996] Davatzikos C, Prince JL, and Bryan RN: Image registration based on boundary mapping. IEEE Trans. Med. Imag., 15(1): 112-115, 1996.
[van den Elsen, 1995] van den Elsen PA, Maintz JBA, Pol EJD, and Viergever MA: Automatic registration of CT and MR brain images using correlation of geometric features. IEEE Trans. Med. Imag., 14(2): 384-396, 1995.
[Enander, 1996] Enander EP, Sjoberg A, Melin B, and Isaksson P: The Matlab Handbook. Addison-Wesley, Reading, Mass., U.S.A., 1996.
[Friston, 1995] Friston KJ, Ashburner J, Frith CD, Poline JB, Heather JD, and Frackowiak RSJ: Spatial registration and normalization of images. Human Brain Mapping, 3: 165-189, 1995.
[Geman, 1984] Geman D and Geman S: Stochastic relaxation, Gibbs distribution bayesian restoration of images. IEEE Trans. Pattern Anal. Machine Intell., 6: 721-741,1984.
[Goodall, 1983] Goodall C: M-Estimators of location: an outline of the theory. In: Understanding Robust and Exploratory Data Analysis, pp. 211-243. Hoaglin DC, F. Mosteller, and Tukey JW Eds. John Willey & Sons, New York, NY, U.S.A., 1983
[Greenstein, 1994] Greenstein OS, Schnall MD, Livolsi VA, and Troupin RH: Suspicious breast lesions: MR imaging with radiologic-pathologic correlation. Radiology, 190: 485-493, 1994.
[Hajnal, 1995] Hajnal JV, Safed N, Soar EJ, Oatridge A, Young IR, and Bydder GM: A registration and interpolation procedure for subvoxel matching of serially acquired MR images. Journal of Computer Assisted Tomography, 19(2): 289-296, 1995.
[Haralick, 1985] Haralick RM and Shapiro LG: Image segmentation techniques. Comput. Vision Graph Image Proc., 29:100-132, 1985.
[Haralick, 1989] Haralick RM, Joo H, Lee CN, Zhuang X, Vaidya V, and Kim MB: Pose estimation from corresponding point data. IEEE Trans. on Systems, Man, and Cybernetics, 19(6): 1426-1446, 1989.
[Haralick, 1992] Haralick RM and Shapiro LG: Computer and Robot Vision, Volume I. Addison-Wesley, Reading, Mass., U.S.A., 1992.
[Hawighorst, 1996] Hawighorst H, Knapstein PG, and Weikel W: Cervical carcinoma: Comparison of standard and pharmacokinetic imaging. Radiology, 201: 531-539, 1996.
[Hogg, 1979] Hogg RJ: An introduction to robust estimation. In: Robustness in Statistics, Launer RL and Wilkinson DN Eds. Academic Press, San Diego, CA, U.S.A., 1979.
[Huber, 1981] Huber PJ: Robust Statistics. John Wiley & Sons, New York, U.S.A., 1981.
[Jiang, 1995] Jiang A, Kennedy DN, Baker JR, Weisskoff RM, Tootel RBH, and Woods RP: Motion detection and correction in functional MR imaging. Human Brain Mapping, 3: 1-12, 1995.
[Jensen, 1986] Jensen JR: Introductory Digital Image Processing: A Remote Sensing Perceptive. Prentice-Hall, Englewood Cliffs, New York, U.S.A., 1986.
[Kass, 1987] Kass M, Witkin A, and Terzopoulos D: Snakes: active contour models. Int. J. Computer Vision, 1(4): 321-331, 1987.
[Lawson, 1995] Lawson W, Ho BT, Shaari CM and Biller HF: Inverted papilloma: a report of 112 cases. Laryngoscope, 105: 282-288, 1995.
[Le, 1996] Le T H and Hu X: Retrospective estimation and correction of physiological artifacts in fMRI by direct extraction of physiological activity from MR data. Magn. Reson. Med., 35: 290-298, 1996.
[Lucht, 2001] Lucht REA, Delorme S, and Brix G: An artificial neural network for the segmentation of Dynamic MR mammographic image series. Proceedings of ISMRM Joint Annual Meeting, Scotland, April 21-27, 839, 2001.
[Maes, 1997] Maes F, Collignon A, Vandermeulen D, Marechal G, and Suetens R: Multimodality image registration by maximization of mutual information. IEEE Transactions on Medical Imaging, 16(2): 187-198, 1997.
[Maintz, 1998] Maintz JBA and Viergever MA: A survey of medical image registration. Medical Image Analysis, 2(1): 1-36, 1998.
[Maurer, 1996] Maurer CR, Aboutanou GB, Dawant BM, Maciunas RJ, and Fitzpatrick JM: Registration of 3-D images using weighted geometrical features. IEEE Trans. Med. Imag., 15(6): 836-849, 1996.
[Meer, 1990] Meer P, Mintz D, Rosenfeld A, and Kim DY: Robust regression methods for computer vision: A review. International Journal of Computer Vision, 6(1): 59-70, 1990.
[Muller-Schimpfle, 1993] Muller-Schimpfle M, Brix G, and Layer G: Recurrent rectal cancer: Diagnosis with dynamic MR imaging. Radiology, 189: 881-889, 1993.
[Nestares, 2000] Nestares O and Heeger DJ: Robust multiresolution alignment of MRI brain volumes. Magnetic Resonance in Medicine, 43: 705-715, 2000.
[Nikou, 1998] Nikou C, Heitz F, Armspach JP, Namer IJ, and Grucker D: Registration of MR/MR and MR/SPECT brain images by fast stochastic optimization of robust voxel similarity measures. NeuroImage, 8: 30-43, 1998.
[Perez, 1994] Perez P and Gidas B: Motion detection and tracking using deformable templates. Proceedings of ICIP, 272-276, 1994.
[Philip, 1994] Philip KP, Dove EL, Mcpherson DD, Gotteiner NL, Standford W, and Chandran KB: The fuzzy hough transform-feature extraction in medical images. IEEE Trans. Med. Imag., 13(2): 235-240, 1994.
[Press, 1998] Press WH, Flannery BP, Teukolsky SA, and Vetterling WT: Numerical Recipes in C, pp. 540-547. Cambridge University Press, Cambridge, U.K., 1988.
[Rajapakse, 1997] Rajapakse JC, Giedd JN, and Rapoport JL: Statistical approach to segmentation of single-chanel cerebral MR images. IEEE Trans. Med. Imag., 16(2): 176-186, 1997.
[Rousseeuw, 1987] Rousseeuw PJ and Leory AM: Robust regression and Outlier Detection. John Wiley & Sons, New York, NY, U.S.A., 1987.
[Shanmugaratnam, 1979] Shanmugaratnam K, Char SN, and de-The G: Histopathology of nasopharyngeal carcinoma: Correlations with epidemiology, survival rates and other biological characteristics. Cancer, 44: 1029-1044, 1979.
[ Studholme, 1997] Studholme C, Hill DLG, and Hawkes DJ: Automated 3D registration of MR and PET brain images by multi-resolution optimization of voxel similarity measures. Medical Physics, 24(1):25-35, 1997.
[Su, 1999] Su MY, Wang Z, Carpenter P, Lao X, Muhler A, and Nalcioglu O: Characterization of N-Ethyl-N-Nitrosourea-induced malignant and benign breast tumors in Rats by using three MR contrast agents. Journal of Magnetic Resonance Imaging, 9: 177-186, 1999.
[Szeliski, 1994] Szeliski R and Weiss R: Robust shape recovery from occluding contours using a linear smoother. Image Understanding Workshop, 939-948, 1994.
[Szeliski , 1996] Szeliski R: Video mosaics for virtual environments. IEEE CG&A, 16: 22-30, 1996.
[Thevenaz, 1998] Thevenaz P, Ruttimann UE, and Unser M: A pyramid approach to subpixel registration based on intensity. IEEE Trans. on Imag. Proc., 7(1): 27-41, 1998.
[Thurfjell, 2000] Thurfjell L, Lau YH, Anderson JLR, and Hutton BF: Improved efficiency for MRI-SPET registration based on mutual information. European Journal of Nuclear Medicine, 7: 847-856, 2000.
[Viola, 1997] Viola P and Wells WM: Alignment by maximization of mutual information. International Journal of Computer Vision, 24(2): 137-154, 1997.
[Woods, 1992] Woods RP, Cherry SR, and Mazziotta JC: Rapid automated algorithm for aligning and reslicing PET images. Journal of Computer Assisted Tomography, 16: 620-633, 1992.
[Woods, 1998] Woods RP, Grafton ST, Holmes CJ, Cherry SR, and Mazziottab JC: Automated image registration: I. General methods and intrasubject, intramodality validation. Journal of Computer Assisted Tomography, 22(1): 139-152, 1998.
[Wu, 1997] Wu DH, Lewin JS, and Duerk JL: Inadequacy of motion correction algorithms in functional MRI: role of susceptibility-induced artefacts. Journal of MRI, 7: 365-370, 1997.
[Yezzi, 1997] Yezzi A, Kichenassamy S, Kumar A, Olver P, and Tannenbaum A: A geometric snake model for segmentation of medical imagery. IEEE Trans. Med. Imag., 16(2): 199-209, 1997.
[Yousem, 1992] Yousem D, Fellows DW, and Kennedy DW: Inverted papilloma evaluation with MR imaging. Radiology, 185: 501-505, 1992.
[Zhu, 1997] Zhu Y and Yan H: Computerized tumor boundary detection using a hopfield neural network. IEEE Trans. Med. Imag., 16(1): 55-67, 1997.