基于主动轮廓模型的脑肿瘤分割技术研究
|
摘要
近20年以来,随着各种新的医学成像技术的临床应用,脑肿瘤诊断和治疗技术取得了很大的进展。磁共振成像(MRI)作为诊断脑肿瘤的一种重要的手段,不仅对软组织具有很高的分辨率,而且对人体无损,因此,当今MRI成了人们进行脑功能、病理和解剖研究的主要手段。肿瘤边界轮廓包含着丰富的肿瘤病变特征信息,临床医生往往是通过对肿瘤病变轮廓特征信息的分析,来对疾病进行定性和分类的。因此,精确的脑组织分割对于探测脑肿瘤的病理类型、放疗计划的制定、外科手术计划的制定和仿真、脑结构的3D可视化和定量测量等具有重要意义。
主动轮廓模型(active contour model, Snake)是由Kass在1987年提出的,是医学图像分割的一种重要工具,该模型的主要原理是先提供待分割图像的一个初始的轮廓位置,并对其定义一个能量函数,使轮廓沿能量降低的方向靠近。当能量函数达到最小时,所提供的初始轮廓收敛到目标的真实轮廓。传统主动轮廓模型存在对初始轮廓线敏感和不能收敛于凹形边缘的问题,尤其是需要依赖其他机制将初始轮廓线放置在感兴趣的图像特征附近。
本文详细介绍了传统主动轮廓模型的算法原理,并从初始轮廓选取和外部力定义两方面对传统主动轮廓模型提出改进:首先,针对传统主动轮廓模型初始轮廓选取时所面临的问题,结合脑肿瘤MRI图像的特点,采用改进的区域增长算法将MRI图像中的肿瘤区域分割出来,并将得到的边界作为主动轮廓模型的初始边缘轮廓;然后分别用sobel梯度算子和梯度矢量流(GVF)进行主动轮廓模型外部力的计算,提高了目标区域的提取精度。与手动选取初始轮廓相比,本文提供的方法可以减少人为干预,提高效率,实验结果表明,改进的主动轮廓模型在脑肿瘤的轮廓提取中能取得良好效果。但是对于浸润性生长的肿瘤,成像中往往与正常组织交叠,边缘模糊,本文中介绍的基于GVF的主动轮廓模型可以比传统的基于梯度的模型更好地辨别分割出真实轮廓。
Since the recent 20 years, the technology of diagnosis and treatment on brain tumor evolved for a high degree, along with the clinical application of various medical imaging technologies. As an important instrument for diagnosis of brain tumor, Magnetic Resonance Imaging (MRI) has high resolving capability for parenchyma, and is without harm to human body, so MRI is the main instrument for research on brain function and pathology nowadays. As the edge contour of tumor includes abounding characteristic information of tumor pathological changes, clinicians always determine the nature of disease by analyzing the characteristic information of tumor contour. So accurate segmentation of brain tissue has important sense for diagnosis and treatment of brain tumor. The active contour model(Snake), originally presented by Kass in 1987, is an important
instrument for medical image segmentation. The main principle of this model is that offering an initial contour with an energy function of the image to be segmented, and push the contour converging along the direction that the energy depresses, and the initial contour converges to the real contour when the energy function arrives at its minimum. The original active contour model has its intrinsic limits: First, the initial contour must, in general, be close to the real boundary; Second, poor convergence to boundary concavities. Especially it need to set initial contour near to the real boundary.
This thesis first introduces the arithmetic principle of original active contour model detailedly, and improves the original model from sides of initial contour selection and definition of external force: first, considering the character of MRI image of brain tumor, adopts improved region grow algorithm for pre-segmentation to get the initial contour of active contour model. Second, use sobel gradient operator and GVF respectively for the calculating of external force of active contour model, therefore, the accuracy of ROI detection is improved. Compared with manual selection of initial contour, the method presented by this thesis can improve the efficiency, The experimental results show that the improved active contour model can get a better result for contour detection of MRI brain tumor, but for soaked tumor that with illegibility boundary, GVF-Snake can get a better result.
主动轮廓模型(active contour model, Snake)是由Kass在1987年提出的,是医学图像分割的一种重要工具,该模型的主要原理是先提供待分割图像的一个初始的轮廓位置,并对其定义一个能量函数,使轮廓沿能量降低的方向靠近。当能量函数达到最小时,所提供的初始轮廓收敛到目标的真实轮廓。传统主动轮廓模型存在对初始轮廓线敏感和不能收敛于凹形边缘的问题,尤其是需要依赖其他机制将初始轮廓线放置在感兴趣的图像特征附近。
本文详细介绍了传统主动轮廓模型的算法原理,并从初始轮廓选取和外部力定义两方面对传统主动轮廓模型提出改进:首先,针对传统主动轮廓模型初始轮廓选取时所面临的问题,结合脑肿瘤MRI图像的特点,采用改进的区域增长算法将MRI图像中的肿瘤区域分割出来,并将得到的边界作为主动轮廓模型的初始边缘轮廓;然后分别用sobel梯度算子和梯度矢量流(GVF)进行主动轮廓模型外部力的计算,提高了目标区域的提取精度。与手动选取初始轮廓相比,本文提供的方法可以减少人为干预,提高效率,实验结果表明,改进的主动轮廓模型在脑肿瘤的轮廓提取中能取得良好效果。但是对于浸润性生长的肿瘤,成像中往往与正常组织交叠,边缘模糊,本文中介绍的基于GVF的主动轮廓模型可以比传统的基于梯度的模型更好地辨别分割出真实轮廓。
Since the recent 20 years, the technology of diagnosis and treatment on brain tumor evolved for a high degree, along with the clinical application of various medical imaging technologies. As an important instrument for diagnosis of brain tumor, Magnetic Resonance Imaging (MRI) has high resolving capability for parenchyma, and is without harm to human body, so MRI is the main instrument for research on brain function and pathology nowadays. As the edge contour of tumor includes abounding characteristic information of tumor pathological changes, clinicians always determine the nature of disease by analyzing the characteristic information of tumor contour. So accurate segmentation of brain tissue has important sense for diagnosis and treatment of brain tumor. The active contour model(Snake), originally presented by Kass in 1987, is an important
instrument for medical image segmentation. The main principle of this model is that offering an initial contour with an energy function of the image to be segmented, and push the contour converging along the direction that the energy depresses, and the initial contour converges to the real contour when the energy function arrives at its minimum. The original active contour model has its intrinsic limits: First, the initial contour must, in general, be close to the real boundary; Second, poor convergence to boundary concavities. Especially it need to set initial contour near to the real boundary.
This thesis first introduces the arithmetic principle of original active contour model detailedly, and improves the original model from sides of initial contour selection and definition of external force: first, considering the character of MRI image of brain tumor, adopts improved region grow algorithm for pre-segmentation to get the initial contour of active contour model. Second, use sobel gradient operator and GVF respectively for the calculating of external force of active contour model, therefore, the accuracy of ROI detection is improved. Compared with manual selection of initial contour, the method presented by this thesis can improve the efficiency, The experimental results show that the improved active contour model can get a better result for contour detection of MRI brain tumor, but for soaked tumor that with illegibility boundary, GVF-Snake can get a better result.
引文
[1] Bloch F, Hansen W.W, Packard M. Nuclear Induction[J]. Phys Rev,1946,69(1):127
[2] Purcell E.M, Torrey H.C, Pround R.V, et al. Resonance absorption by nuclear magnetic moments in a solid[J]. Phys Rev,1946,69(1):37～38
[3] Lauterbur P C. Image formation by induced local interactions: Examples employing nuclear magnetic resonance[J]. Nature,1973,242:190～191
[4]曲德鑫.磁共振成像MRI的原理及发展动态[J].医学放射技术杂志,2005,(7):15～16
[5] Prastawa M, Bullitt E. Automatic brain tumor segmentation by subject specific modification of atlas priors[J]. Medical Image Computing, Academic Radiology, 2003,10(12):1341～1348
[6] Kass M, Witkin A, Terzopoulous D. Snake:active counter models. In:Brady I M,Rosenfield A eds. Proceedings of the 1st International Conference on Computer Vision. London: IEEE Computer Society Press,1987.259～268
[7] Amini A A, Weymouth T E, Jain R C. Using dynamic programming for solving variational problem in vision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1990,12(9):855～867
[8] Williams D J, Shab M. A fast algorithm for active contours and curvature estimation[J]. CVGIP: Image Understanding,1992,55(1):14～26
[9] Kass M, Witkin A, TerzoPoulous D. Snakes: Active Contour Models. International Journal of Computer Vision,1988,1(4):321～331
[10] Xu C, Prince J L. Snakes, shapes, and gradient vector flow[J]. IEEE Trans Image Proc, 1998,7(3):359～369
[11]邓航,余松煜.综合利用通用霍夫变换与Snake算法对序列图像的分割[J].红外与激光工程,2000,29(2):9～11
[12]成金勇,范延滨,宋洁等.基于小波分析与Snake模型的图像边缘检测方法[J].青岛大学学报,2005,18(1):77～81
[13]王蓓,张立明.利用图像先验知识与snake结合对心脏序列图像的分割[J].复旦学报,2003,42(1):81～86
[14]李涛,于明,兰娜.一种自动初始化轮廓的改进蛇模型算法[J].河北工业大学学报,2006,35(1):58～61
[15]邱明,张二虎,张志刚.基于改进的GVF模型的CT图像分割方法[J].小型微型计算机系统,2006,27(1):155～157
[16] Thedens D R. Methods of graph searching for border detection in image sequence with application to cardiac images[J]. IEEE Transaction Medical Imaging,1995, 14(3):42～55
[17] Ivana Mikic. Segmentation and tracking in echocardiography sequences: active contours guided by optical flow estimates[J]. IEEE Transaction on Medical Imaging, 1998,17(2):274～284
[18]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.1～2
[19]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.179～180
[20]章毓晋.图像分割. (第一版).北京:科学出版社, 2001.2.43～49
[21] Grimson W E L, Perez T L. Localizing overlapping parts by searching the interpretation tree[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1987,9:469～482
[22]章毓晋.图像分割.(第一版).北京:科学出版社, 2001.2.67～73
[23] Dzang L Pham, Jerry L Prince. An adaptive fuzzy C means algorithm for image segmentation in the presence of intensity inhomogeneities[J]. Pattern Recognition Letters,1999,20:57～68
[24]吴林,郭大勇,施克仁等.改进的FCM在人脑MR图像分割中的应用[J].清华大学学报(自然科学版),2004,42(2):157～159
[25]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.182～184
[26] J Canny. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1986,8(6):679～696
[27] Vishvjit S. Nalwa, Thomas O.Binford. On Detecting Edges[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,1986,8(6):699～714
[28] Robert M.Haralick, FELLOW, IEEE. Digital Step Edges from Zero Crossing of Second Directional Derivatives[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,1984,6(1):58～68
[29] Staib L H, Duncan J S. Boundary finding with parametrically deformable models[J].IEEE Trans.PAMI,1992,14(11):1061～1075
[30] Falcao Alexandre X, Udupa Jayaram K. User-Steered Image Segmentation Paradigms:Live Wire and Live Lane[J]. Graphical Models and Image Processing, 1998, 60(4):233～260
[31] F Meyer, S beucher. Morphology segmentation[J]. J Visual Comm and Image Representation,1990,1(1):21～26
[32]劳丽,吴效明,朱学峰.模糊集理论在图像分割中的应用综述[J].中国体视学与图像分析,2006,11(3):200～205
[33]陈果,左洪福.图像的自适应模糊阈值分割法[J].自动化学报,2003,29(5): 791～796
[34] Udupa J K, Samarasekera S. Fuzzy Connectedness and Object Definition:Theory, Algorithms, and Application in image Segmentation[J]. Graphical Models and Image Processing,1996,58(3):246～261
[35] Blanz WE, Gish SL. A connectionist classifier architecture applied to image segmentation. In:Proc.10th ICPR, International Conference on Pattern Recognition. London:IAPR,1990,272～277
[36] Babaguchi N, Yamada K. Connectionist model binarization. In:Proc.10th ICPR, International Conference on Pattern Recognition.London:IAPR,1990,51～56
[37]贺士娟,李颖.基于可变形模型算法提取MRI图像脑区域的方法[J].河北工业大学学报,2002,31(2):1～5
[38]徐牧,王润生.一种组合主动轮廓线模型算法[J].计算机工程与科学, 2004,26(12):38～41
[39]王立功,于甬华.基于snake模型的图像目标轮廓自动跟踪方法[J].东南大学学报, 2003,33(2):215～218
[40]刘彩霞,范延滨. GVF snake模型中一种新的初始轮廓设置方法[J].计算机应用, 2006,26(7):1614～1616
[41]李丽勤,高焕文,周兴祥. Snake模型初始轮廓选取的研究[J].计算机工程与应用,2004,11:43～45
[42]金雪军,蔡家楣,冯晓斐. Snake初始模型及其改进算法的研究[J].浙江工业大学学报,2006,34(2):166～169
[2] Purcell E.M, Torrey H.C, Pround R.V, et al. Resonance absorption by nuclear magnetic moments in a solid[J]. Phys Rev,1946,69(1):37～38
[3] Lauterbur P C. Image formation by induced local interactions: Examples employing nuclear magnetic resonance[J]. Nature,1973,242:190～191
[4]曲德鑫.磁共振成像MRI的原理及发展动态[J].医学放射技术杂志,2005,(7):15～16
[5] Prastawa M, Bullitt E. Automatic brain tumor segmentation by subject specific modification of atlas priors[J]. Medical Image Computing, Academic Radiology, 2003,10(12):1341～1348
[6] Kass M, Witkin A, Terzopoulous D. Snake:active counter models. In:Brady I M,Rosenfield A eds. Proceedings of the 1st International Conference on Computer Vision. London: IEEE Computer Society Press,1987.259～268
[7] Amini A A, Weymouth T E, Jain R C. Using dynamic programming for solving variational problem in vision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1990,12(9):855～867
[8] Williams D J, Shab M. A fast algorithm for active contours and curvature estimation[J]. CVGIP: Image Understanding,1992,55(1):14～26
[9] Kass M, Witkin A, TerzoPoulous D. Snakes: Active Contour Models. International Journal of Computer Vision,1988,1(4):321～331
[10] Xu C, Prince J L. Snakes, shapes, and gradient vector flow[J]. IEEE Trans Image Proc, 1998,7(3):359～369
[11]邓航,余松煜.综合利用通用霍夫变换与Snake算法对序列图像的分割[J].红外与激光工程,2000,29(2):9～11
[12]成金勇,范延滨,宋洁等.基于小波分析与Snake模型的图像边缘检测方法[J].青岛大学学报,2005,18(1):77～81
[13]王蓓,张立明.利用图像先验知识与snake结合对心脏序列图像的分割[J].复旦学报,2003,42(1):81～86
[14]李涛,于明,兰娜.一种自动初始化轮廓的改进蛇模型算法[J].河北工业大学学报,2006,35(1):58～61
[15]邱明,张二虎,张志刚.基于改进的GVF模型的CT图像分割方法[J].小型微型计算机系统,2006,27(1):155～157
[16] Thedens D R. Methods of graph searching for border detection in image sequence with application to cardiac images[J]. IEEE Transaction Medical Imaging,1995, 14(3):42～55
[17] Ivana Mikic. Segmentation and tracking in echocardiography sequences: active contours guided by optical flow estimates[J]. IEEE Transaction on Medical Imaging, 1998,17(2):274～284
[18]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.1～2
[19]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.179～180
[20]章毓晋.图像分割. (第一版).北京:科学出版社, 2001.2.43～49
[21] Grimson W E L, Perez T L. Localizing overlapping parts by searching the interpretation tree[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1987,9:469～482
[22]章毓晋.图像分割.(第一版).北京:科学出版社, 2001.2.67～73
[23] Dzang L Pham, Jerry L Prince. An adaptive fuzzy C means algorithm for image segmentation in the presence of intensity inhomogeneities[J]. Pattern Recognition Letters,1999,20:57～68
[24]吴林,郭大勇,施克仁等.改进的FCM在人脑MR图像分割中的应用[J].清华大学学报(自然科学版),2004,42(2):157～159
[25]章毓晋.图像工程(上册)-图像处理和分析.(第一版).北京:清华大学出版社, 1999.2.182～184
[26] J Canny. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1986,8(6):679～696
[27] Vishvjit S. Nalwa, Thomas O.Binford. On Detecting Edges[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,1986,8(6):699～714
[28] Robert M.Haralick, FELLOW, IEEE. Digital Step Edges from Zero Crossing of Second Directional Derivatives[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,1984,6(1):58～68
[29] Staib L H, Duncan J S. Boundary finding with parametrically deformable models[J].IEEE Trans.PAMI,1992,14(11):1061～1075
[30] Falcao Alexandre X, Udupa Jayaram K. User-Steered Image Segmentation Paradigms:Live Wire and Live Lane[J]. Graphical Models and Image Processing, 1998, 60(4):233～260
[31] F Meyer, S beucher. Morphology segmentation[J]. J Visual Comm and Image Representation,1990,1(1):21～26
[32]劳丽,吴效明,朱学峰.模糊集理论在图像分割中的应用综述[J].中国体视学与图像分析,2006,11(3):200～205
[33]陈果,左洪福.图像的自适应模糊阈值分割法[J].自动化学报,2003,29(5): 791～796
[34] Udupa J K, Samarasekera S. Fuzzy Connectedness and Object Definition:Theory, Algorithms, and Application in image Segmentation[J]. Graphical Models and Image Processing,1996,58(3):246～261
[35] Blanz WE, Gish SL. A connectionist classifier architecture applied to image segmentation. In:Proc.10th ICPR, International Conference on Pattern Recognition. London:IAPR,1990,272～277
[36] Babaguchi N, Yamada K. Connectionist model binarization. In:Proc.10th ICPR, International Conference on Pattern Recognition.London:IAPR,1990,51～56
[37]贺士娟,李颖.基于可变形模型算法提取MRI图像脑区域的方法[J].河北工业大学学报,2002,31(2):1～5
[38]徐牧,王润生.一种组合主动轮廓线模型算法[J].计算机工程与科学, 2004,26(12):38～41
[39]王立功,于甬华.基于snake模型的图像目标轮廓自动跟踪方法[J].东南大学学报, 2003,33(2):215～218
[40]刘彩霞,范延滨. GVF snake模型中一种新的初始轮廓设置方法[J].计算机应用, 2006,26(7):1614～1616
[41]李丽勤,高焕文,周兴祥. Snake模型初始轮廓选取的研究[J].计算机工程与应用,2004,11:43～45
[42]金雪军,蔡家楣,冯晓斐. Snake初始模型及其改进算法的研究[J].浙江工业大学学报,2006,34(2):166～169