一种基于眼底图像的视网膜血管三维重建方法
摘要
眼底血管是诊病的重要依据，本文根据眼球的解剖学特征进行分析，建立了适于医学可视化的眼底数学模型。并利用中心扩散逆投影法重建了眼底曲面，由原始的彩色图像经过图像类型转换、图像增强、滤波、图像填充、边缘提取得到了血管轮廓图、骨架图和边缘图，从血管骨架图中查找血管，利用轮廓图确定血管的半径，使用边缘图分析出血管交叉处的层次关系，并最终完成了血管的三维重建和显示。针对眼底血管系统提出了血管关键点选取算法、骨架修正、血管交叉点识别算法、血管层次识别等算法，并且使用了部分技巧，使算法不仅达到了较高的精度，增强血管清晰程度。为临床应用提供了三维眼底血管的参数采集和显示环境，为诊断、显示、激光治疗、手术定位等提供有益的帮助。
The development of computer graphics and image technology has contributed to wide application of 3D medical imaging technology in medical diagnosis and treatment, as is the research focus at present. 3D medical images provide strong evidence for clinic diagnosis, therapy options, curative effects investigation, prognosis measurement and pathogenic mechanism exploration. They also play an enormous role in operation planning and simulation, anatomy, education and medical research. Ocular fundus images are an objective standard for diagnosis. Many diseases, such as diabetes, leukaemia, hypertension, arteriosclerosis and the like, can be diagnosed and treated using the images. It has been proved that application of 3D ocular fundus images makes medical examination sharper and more effective, which provides evidence for clinic diagnosis, therapy options, curative effects investigation, prognosis measurement and pathogenic mechanism exploration. Based on those former researches, this thesis brings forward a way of 3D reconstruction of retina veins and puts it to realization, and it will be a practical guide for future project of ocular fundus images processing.
First, the thesis discusses image-based modeling technique, a technique that uses images to reconstruct geometric models of an object. According to the amount of images resources required for modeling, modeling can be divided into the single-imaged and the multi-imaged. This thesis is aiming at a specific issue of single-imaged modeling. The two existing means of single-imaged reconstruction of 3D models are as follows: one is the introduction of repository to make full use of accumulated geometric knowledge on reconstruction; the other is to draw in more interaction, namely, the users directly indicate relevant information on images. A typical problem incurred in single-imaged modeling lies in how to restore the physical shapes through chiaroscuro. Its principle is to analyze the vector of the physical surface through the brightness variety on the surface of images, and then to translate it into deep information. Compared with traditional modeling technique, image-based modeling technology is easier, quicker, and more real. Having been extensively applied, the technology is intended to develop more rapidly.
Image segmentation is an important component of research on 3D image reconstruction and visualization. The fundus blood vessels are so complicated that only when they are detected can people make a purposeful and selective analysis and display them. After pre-processed, the blood vessels on the images can be separated from other objects and then be used for 3D reconstruction. First, to change color image to grayscale, and then to enforce images. Enforcing images is to strengthen contrast of the images by use if grayscale modification and histogram modification. The main purpose is to make fundus blood vessels appear sharper on the processed images, so as to make better visual effects. Image filtering will remove image noise to fulfill binalization of final images. This thesis briefly introduces the working principle of linear filter, median filter and Wiener filter.
Hole filling is able to give removal treatment to some small areas of the filtered binal images. After filling some irrelevant irregular areas around macular area and blood vessels can be removed, so is done to some white disturbing areas. Skeleton detection is to use the corrosion algorithm to transform the vessel contour into vessel skeleton. As regards vessel boundary extraction, the thesis introduces the principles of Roberts arithmetic operator, Sobel operator and Prewitt operator, with focus on how to apply edge map created by edge operator to process the layers of vessels.
Detection of fundus blood vessels is key to vessel modeling. It is mainly using the skeleton map to find out vessel alteration, and using contour map to calculate radius of vessel, and analyzing the intersection of the distributed vessels according to edge map. Eyeball is oval, so we can produce the arithmetic formula about the eyeballs and fundus surfaces. By this means, we can get the mapping relations between any spot in 2D fundus image and its corresponding spot in 3D fundus surface, as well as the theoretic support for 3D reconstruction of fundus vessels. By the use of skeleton map, we can put the vessels into several segments for easier management in the future. Utilizing key points, we can give a concise and correct relation to vessel distribution, since those points can outline its basic distribution.
Since the skeleton map is formed out of the contour map by means of the corrosion algorithm, the points on the skeleton map can not completely match with the axis of the vessel tree, therefore, it is incorrect to build a vessel model with that as the center line of vessel. We need to make alteration to the key points to make them closer to the axis; meanwhile, the determination of key points presented in this thesis is to adjust key points to the axis of the vessel, and calculate the radius of the vessel. The edge map of the vessels enable us to find out the hierarchy information of vessels at the intersection. This thesis also presents the application of intersection detection to distinguish vessel embranchment from vessel intersection, the upper layer of vessels from the lower layer by use of edge map. It also provides variance algorithm to detect layers of vessels by using fundus grayscale map. To date, we can go on with 3D reconstruction with all these data, including key points, radius of vessels at key points, intersection, the layers of intersection.
The first requirement is to calculate the section through the key point and perpendicular to vessel axis. We simply require the section be perpendicular to the flat surface which is perpendicular to our lines of sight, instead of the one exactly perpendicular to the axis. The effect remains the same. Now use the key point as the centre, the radius of vessel as the radius, to make a circle in the section, and the circular section is formed of the vessel wall at key point and the section. With the circles on both ends of a vessel segment, we can outline this vessel segment in form of column vessel. Linking all these segments up, we construct 3D model of the whole retina vein system. We use D3D software of Microsoft to demonstrate vessel system. 3D images are displayed after setting up lights on the location and determining the materials used. By the set-up of coordinate system, we can switch angles to realize 3D display, so that the users can make an all-round observance of the condition of vessels and their distribution.
The 3D model of vessels in this thesis can demonstrate clearly the 3D images of fundus vessles. Certainly, it is far from perfect. For example, we can measure the vessel parameter based on the 3D images of vessels and then analyze the alteration of fundus micro-vessels, which is useful for the diagnosis of such diseases as hypertension, heart diseases and all the others that can lead to alteration in vessel diameter.
The development of computer graphics and image technology has contributed to wide application of 3D medical imaging technology in medical diagnosis and treatment, as is the research focus at present. 3D medical images provide strong evidence for clinic diagnosis, therapy options, curative effects investigation, prognosis measurement and pathogenic mechanism exploration. They also play an enormous role in operation planning and simulation, anatomy, education and medical research. Ocular fundus images are an objective standard for diagnosis. Many diseases, such as diabetes, leukaemia, hypertension, arteriosclerosis and the like, can be diagnosed and treated using the images. It has been proved that application of 3D ocular fundus images makes medical examination sharper and more effective, which provides evidence for clinic diagnosis, therapy options, curative effects investigation, prognosis measurement and pathogenic mechanism exploration. Based on those former researches, this thesis brings forward a way of 3D reconstruction of retina veins and puts it to realization, and it will be a practical guide for future project of ocular fundus images processing.
First, the thesis discusses image-based modeling technique, a technique that uses images to reconstruct geometric models of an object. According to the amount of images resources required for modeling, modeling can be divided into the single-imaged and the multi-imaged. This thesis is aiming at a specific issue of single-imaged modeling. The two existing means of single-imaged reconstruction of 3D models are as follows: one is the introduction of repository to make full use of accumulated geometric knowledge on reconstruction; the other is to draw in more interaction, namely, the users directly indicate relevant information on images. A typical problem incurred in single-imaged modeling lies in how to restore the physical shapes through chiaroscuro. Its principle is to analyze the vector of the physical surface through the brightness variety on the surface of images, and then to translate it into deep information. Compared with traditional modeling technique, image-based modeling technology is easier, quicker, and more real. Having been extensively applied, the technology is intended to develop more rapidly.
Image segmentation is an important component of research on 3D image reconstruction and visualization. The fundus blood vessels are so complicated that only when they are detected can people make a purposeful and selective analysis and display them. After pre-processed, the blood vessels on the images can be separated from other objects and then be used for 3D reconstruction. First, to change color image to grayscale, and then to enforce images. Enforcing images is to strengthen contrast of the images by use if grayscale modification and histogram modification. The main purpose is to make fundus blood vessels appear sharper on the processed images, so as to make better visual effects. Image filtering will remove image noise to fulfill binalization of final images. This thesis briefly introduces the working principle of linear filter, median filter and Wiener filter.
Hole filling is able to give removal treatment to some small areas of the filtered binal images. After filling some irrelevant irregular areas around macular area and blood vessels can be removed, so is done to some white disturbing areas. Skeleton detection is to use the corrosion algorithm to transform the vessel contour into vessel skeleton. As regards vessel boundary extraction, the thesis introduces the principles of Roberts arithmetic operator, Sobel operator and Prewitt operator, with focus on how to apply edge map created by edge operator to process the layers of vessels.
Detection of fundus blood vessels is key to vessel modeling. It is mainly using the skeleton map to find out vessel alteration, and using contour map to calculate radius of vessel, and analyzing the intersection of the distributed vessels according to edge map. Eyeball is oval, so we can produce the arithmetic formula about the eyeballs and fundus surfaces. By this means, we can get the mapping relations between any spot in 2D fundus image and its corresponding spot in 3D fundus surface, as well as the theoretic support for 3D reconstruction of fundus vessels. By the use of skeleton map, we can put the vessels into several segments for easier management in the future. Utilizing key points, we can give a concise and correct relation to vessel distribution, since those points can outline its basic distribution.
Since the skeleton map is formed out of the contour map by means of the corrosion algorithm, the points on the skeleton map can not completely match with the axis of the vessel tree, therefore, it is incorrect to build a vessel model with that as the center line of vessel. We need to make alteration to the key points to make them closer to the axis; meanwhile, the determination of key points presented in this thesis is to adjust key points to the axis of the vessel, and calculate the radius of the vessel. The edge map of the vessels enable us to find out the hierarchy information of vessels at the intersection. This thesis also presents the application of intersection detection to distinguish vessel embranchment from vessel intersection, the upper layer of vessels from the lower layer by use of edge map. It also provides variance algorithm to detect layers of vessels by using fundus grayscale map. To date, we can go on with 3D reconstruction with all these data, including key points, radius of vessels at key points, intersection, the layers of intersection.
The first requirement is to calculate the section through the key point and perpendicular to vessel axis. We simply require the section be perpendicular to the flat surface which is perpendicular to our lines of sight, instead of the one exactly perpendicular to the axis. The effect remains the same. Now use the key point as the centre, the radius of vessel as the radius, to make a circle in the section, and the circular section is formed of the vessel wall at key point and the section. With the circles on both ends of a vessel segment, we can outline this vessel segment in form of column vessel. Linking all these segments up, we construct 3D model of the whole retina vein system. We use D3D software of Microsoft to demonstrate vessel system. 3D images are displayed after setting up lights on the location and determining the materials used. By the set-up of coordinate system, we can switch angles to realize 3D display, so that the users can make an all-round observance of the condition of vessels and their distribution.
The 3D model of vessels in this thesis can demonstrate clearly the 3D images of fundus vessles. Certainly, it is far from perfect. For example, we can measure the vessel parameter based on the 3D images of vessels and then analyze the alteration of fundus micro-vessels, which is useful for the diagnosis of such diseases as hypertension, heart diseases and all the others that can lead to alteration in vessel diameter.
引文
[1] 廖琪梅 杭洽时 钱宗才 三维医学图像重建的方法与应用 现代电子技术1997 3 41-42
[2] 李万祥 苟向锋 由切片重建样本的三维形态 信息技术2002 3 14-15
[3] 刘钢 彭群生 鲍虎军 基于图像建模技术研究综述与展望 计算机辅助设计与图形学学报2005 1第17卷311-326
[4] Horry Y, Anjyo K, Arai K. Tour into the picture: using a spidery mesh interface to make animation from a single image. In: Proceedings of the SIGGRAPH'97. New York: ACM Press/Addison-Wesley Publishing Company, 1997, 225-232
[5] Blanz V, Vetter T. A morphable model for the synthesis of 3D faces. In: Proceedings of the SIGGRAPH' 99. New York: ACM Press, 1999, 187-194
[6] Zhang R, Tsai P S, Cryer J E, et al. Shape from shading: a survey. IEEE Transactions on Analysis and Machine Intelligence, 1999, 21(8): 690-705
[7] 赵欲波 张定华 M．Petrou赵荣椿 基于分形的从明暗恢复形状算法研究 中国图像图形学报2002 5第7卷(A)440-444
[8] Scharstein D, Szeliski R. A taxonomy and evaluation of dense two-frame stereo correspondence algorithms. International Journal of Computer Vision, 2002, 47(1)
[9] 杨瑞元 邱建雄 基于图像的建模和绘制技术综述 计算机辅助设计与图形学学报2002 2第14卷186-188
[10] T-L Ji, Sundareshan M K, Roehrig H. Adaptive image contrast enhancement based on human visual properties [J]. IEEE Trans Med Imaging, 1994, 13: 573-586
[11] 何斌 马天予 王运坚 朱红莲 Visum c++数字图像处理 人民邮电 出版社2002年第2版262-282
[12] 潘晨 顾峰 基于3D直方图的彩色图象分割方法 中国图像图形学报2002 8第7卷(A版)800-805
[13] Huang C C, Li C C, Fan N, et al. A fast morphological filter for enhancement of angiographic images [J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1991, 13(1): 0229-0230
[14] 魏巍MATLAB信息工程工具箱技术手册 国防工业出版社2004年第一版12-85
[15] Peng J, Rusch Ph. Morphological filters and edge detection application to medical imaging [J]. Annual International Conference of the IEEE Engineering In Medicine and Biology Society, 1991, 13 (1): 0251-0252.
[16] L. G. Roberts, Machine Perception of Three-Dimension Solids, in Optical and Electro-Optimal Information Processing, J. t. Tippett, et al, Ed. Cambridge, MA: MIT Press, 1965, 99-159-197
[17] 李超 梁斌 陈武凡 吴德正 由二维眼底正投影图像向三维曲面逆投影成像的重建算法 中国生物医学工程学报2002 8第21卷第4期346-350
[18] 王金鹤 扫描图象曲线轮廓关键点的提取及其处理 中国图像图形学报2001 7第6卷(A版)699-702
[19] 傅晓烨 田景成 庞云阶 一种眼底图像的三维重建方法 中国图像图形学报1997 2第7卷509-511
[20] Tascini G, Passerini G, Puliti P, et al. Retina vascular network recognition [J]. Proc SPIE, 1993, 1898: 322-329
[21] 尚晶晶Direct3D游戏开发技术详解人民邮电出版社2006年第1版142-146
[22] A. Can, H. Shen, J. N. T urner, J. L. T anenbaum, and B. Ro ysam, "Rapid automatedtracing and feature extraction from retinal fundus images using direct exploratoryalgorithms," IEEE Transactions on Information Technology in Biomedicine, vol. 3, no. 2, pp. 125-137, June 1999
[23] Huang C C, Li C C, Fan N, et al. A fast morphological filter for enhancement of angiographic images [J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1991, 13(1): 0229-0230
[24] Zascini G, Passerini G, Puliti P, et al. Retina vascular network recognition [J]. Proc SPIE, 1993, 1898:3Z2-329
[25] Chauduri S, Chatterjee S, Katz N, et al. Detection of blood vessels in retinal images using two-dimensional matched filters [J]. IEEE Trans Med Imaging, 1989,8:Z63-Z69
[26] T-L Ji, Sundareshan M K, Roehrig H. Adaptive image contrast enhancement based on human visual properties [J]. IEEE Trans Med Imaging, 1994,13:573-586
[27] 李健 庞云阶著 一种提取医疗图像中血管轮廓的简单方法 系统仿真学报2001 11第13卷增刊561-563
[28] 王晓丹 吴崇明 基于MATLAB的系统分析与设计 西安电子科技大学出版社2000年第1版
[29] 苏彦华等Visual c++数字图像识别技术典型案例 人民邮电出版社2004年第一版
[30] 王福昌 曹慧荣 利用matlab进行血管基于表面的三维重建 电脑学习2002 4 6-7
[31] 张海棠Visual c++编程指南 航空工业出版社1999年第1版
[32] George Shepherd David Kruglinski Visual c++．NET清华大学出版社2004年第1版
[33] 荣钦科技Visual C++游戏设计 北京科海电子出版社2003年第1版
[2] 李万祥 苟向锋 由切片重建样本的三维形态 信息技术2002 3 14-15
[3] 刘钢 彭群生 鲍虎军 基于图像建模技术研究综述与展望 计算机辅助设计与图形学学报2005 1第17卷311-326
[4] Horry Y, Anjyo K, Arai K. Tour into the picture: using a spidery mesh interface to make animation from a single image. In: Proceedings of the SIGGRAPH'97. New York: ACM Press/Addison-Wesley Publishing Company, 1997, 225-232
[5] Blanz V, Vetter T. A morphable model for the synthesis of 3D faces. In: Proceedings of the SIGGRAPH' 99. New York: ACM Press, 1999, 187-194
[6] Zhang R, Tsai P S, Cryer J E, et al. Shape from shading: a survey. IEEE Transactions on Analysis and Machine Intelligence, 1999, 21(8): 690-705
[7] 赵欲波 张定华 M．Petrou赵荣椿 基于分形的从明暗恢复形状算法研究 中国图像图形学报2002 5第7卷(A)440-444
[8] Scharstein D, Szeliski R. A taxonomy and evaluation of dense two-frame stereo correspondence algorithms. International Journal of Computer Vision, 2002, 47(1)
[9] 杨瑞元 邱建雄 基于图像的建模和绘制技术综述 计算机辅助设计与图形学学报2002 2第14卷186-188
[10] T-L Ji, Sundareshan M K, Roehrig H. Adaptive image contrast enhancement based on human visual properties [J]. IEEE Trans Med Imaging, 1994, 13: 573-586
[11] 何斌 马天予 王运坚 朱红莲 Visum c++数字图像处理 人民邮电 出版社2002年第2版262-282
[12] 潘晨 顾峰 基于3D直方图的彩色图象分割方法 中国图像图形学报2002 8第7卷(A版)800-805
[13] Huang C C, Li C C, Fan N, et al. A fast morphological filter for enhancement of angiographic images [J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1991, 13(1): 0229-0230
[14] 魏巍MATLAB信息工程工具箱技术手册 国防工业出版社2004年第一版12-85
[15] Peng J, Rusch Ph. Morphological filters and edge detection application to medical imaging [J]. Annual International Conference of the IEEE Engineering In Medicine and Biology Society, 1991, 13 (1): 0251-0252.
[16] L. G. Roberts, Machine Perception of Three-Dimension Solids, in Optical and Electro-Optimal Information Processing, J. t. Tippett, et al, Ed. Cambridge, MA: MIT Press, 1965, 99-159-197
[17] 李超 梁斌 陈武凡 吴德正 由二维眼底正投影图像向三维曲面逆投影成像的重建算法 中国生物医学工程学报2002 8第21卷第4期346-350
[18] 王金鹤 扫描图象曲线轮廓关键点的提取及其处理 中国图像图形学报2001 7第6卷(A版)699-702
[19] 傅晓烨 田景成 庞云阶 一种眼底图像的三维重建方法 中国图像图形学报1997 2第7卷509-511
[20] Tascini G, Passerini G, Puliti P, et al. Retina vascular network recognition [J]. Proc SPIE, 1993, 1898: 322-329
[21] 尚晶晶Direct3D游戏开发技术详解人民邮电出版社2006年第1版142-146
[22] A. Can, H. Shen, J. N. T urner, J. L. T anenbaum, and B. Ro ysam, "Rapid automatedtracing and feature extraction from retinal fundus images using direct exploratoryalgorithms," IEEE Transactions on Information Technology in Biomedicine, vol. 3, no. 2, pp. 125-137, June 1999
[23] Huang C C, Li C C, Fan N, et al. A fast morphological filter for enhancement of angiographic images [J]. Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1991, 13(1): 0229-0230
[24] Zascini G, Passerini G, Puliti P, et al. Retina vascular network recognition [J]. Proc SPIE, 1993, 1898:3Z2-329
[25] Chauduri S, Chatterjee S, Katz N, et al. Detection of blood vessels in retinal images using two-dimensional matched filters [J]. IEEE Trans Med Imaging, 1989,8:Z63-Z69
[26] T-L Ji, Sundareshan M K, Roehrig H. Adaptive image contrast enhancement based on human visual properties [J]. IEEE Trans Med Imaging, 1994,13:573-586
[27] 李健 庞云阶著 一种提取医疗图像中血管轮廓的简单方法 系统仿真学报2001 11第13卷增刊561-563
[28] 王晓丹 吴崇明 基于MATLAB的系统分析与设计 西安电子科技大学出版社2000年第1版
[29] 苏彦华等Visual c++数字图像识别技术典型案例 人民邮电出版社2004年第一版
[30] 王福昌 曹慧荣 利用matlab进行血管基于表面的三维重建 电脑学习2002 4 6-7
[31] 张海棠Visual c++编程指南 航空工业出版社1999年第1版
[32] George Shepherd David Kruglinski Visual c++．NET清华大学出版社2004年第1版
[33] 荣钦科技Visual C++游戏设计 北京科海电子出版社2003年第1版