一种用于超声图像序列分割的水平集演化方法
|
摘要
超声图像具有低信噪比、边界模糊、边界部分缺失、灰度不均等特点,对它的分割极具挑战性.而图像分割又是图像定量、定性分析的关键环节,分割的精确性对后续的分析、处理工作影响重大.距离保持水平集演化(DRLSE)方法对超声图像中出现的弱边界、被部分遮挡边界的分割较差,容易受噪声和灰度不均的影响,因此易造成弱边界泄漏、局部最优等误分割问题;并且初始轮廓对位置敏感,这使得分割的正确性严重依赖初始轮廓位置的选择,故不能对图像进行批量处理.为此提出了一种优化策略:融合基于局部区域的灰度信息和基于边缘的梯度信息构造新的边缘停止函数和面积项权系数,使得演化曲线不仅能够自适应地改变演化方向更有利于对图像序列的处理,同时对斑点噪声和灰度不均问题也有很好的抑制能力;另外,构造了一个先验形状约束项,利用前一帧的分割结果对当前帧的分割进行约束,促进曲线正确演化至目标边界,使得对边界部分遮挡的图像也有着更精确的分割效果.通过合成图像和真实超声图像对分割算法进行了性能分析,设计了基于边缘的豪斯多夫距离和平均绝对距离对算法分割轮廓和医生分割轮廓之间的距离差异性进行度量,实验证明优化策略相比于DRLSE模型和其传统优化模型,有着更高的分割精度,分割效果更出色.
Due to low signal-noise ratio,blurry boundaries,partially occluded boundaries,and intensity inhomogeneity,ultrasound image segmentation is quite challenging. Image segmentation plays a vital role in quantitative and qualitative analysis of ultrasonic images,and the accuracy of segmentation has a great influence on the subsequent processing work. Distance regularized level set evolution(DRLSE)has a poor segmentation effect on the weak boundary and partially occluded boundary appearing in the ultrasonic image,and it is very sensitive to image noise and intensity inhomogeneity. It is therefore easy to cause weak boundary leakage,local optimum,and other missegmentation. In addition,the DRLSE is highly dependent on initial contour positions,and hence,the image cannot be batch processed. In view of the above defects,a new edge-stop function and weighting coefficient-of-area term have been defined. Based on local grey scale and edge gradient,this new function and term have been devised so that the evolution curve could not only adaptively change the evolution direction,but also facilitate the processing of image sequences,while effectively suppressing noise and intensity inhomogeneity. A prior shape constraint was constructed to constrain the segmentation of the current frame via the segmentation result of the previous frame. The curve was correctly evolved to the target boundary so that the image partially-occluded boundary also displayed a more precise segmentation effect. The performance of the segmentation algorithm was analyzed by synthetic image and real ultrasound image. The edge-based Hausdorff distance and the mean absolute distance were designed to measure the distance difference between the algorithm's segmentation contour and the doctor's segmentation contour. The experimental results showed that,compared with DRLSE and some optimization models of DRLSE,the proposed method improves both segmentation accuracy and effect.
Due to low signal-noise ratio,blurry boundaries,partially occluded boundaries,and intensity inhomogeneity,ultrasound image segmentation is quite challenging. Image segmentation plays a vital role in quantitative and qualitative analysis of ultrasonic images,and the accuracy of segmentation has a great influence on the subsequent processing work. Distance regularized level set evolution(DRLSE)has a poor segmentation effect on the weak boundary and partially occluded boundary appearing in the ultrasonic image,and it is very sensitive to image noise and intensity inhomogeneity. It is therefore easy to cause weak boundary leakage,local optimum,and other missegmentation. In addition,the DRLSE is highly dependent on initial contour positions,and hence,the image cannot be batch processed. In view of the above defects,a new edge-stop function and weighting coefficient-of-area term have been defined. Based on local grey scale and edge gradient,this new function and term have been devised so that the evolution curve could not only adaptively change the evolution direction,but also facilitate the processing of image sequences,while effectively suppressing noise and intensity inhomogeneity. A prior shape constraint was constructed to constrain the segmentation of the current frame via the segmentation result of the previous frame. The curve was correctly evolved to the target boundary so that the image partially-occluded boundary also displayed a more precise segmentation effect. The performance of the segmentation algorithm was analyzed by synthetic image and real ultrasound image. The edge-based Hausdorff distance and the mean absolute distance were designed to measure the distance difference between the algorithm's segmentation contour and the doctor's segmentation contour. The experimental results showed that,compared with DRLSE and some optimization models of DRLSE,the proposed method improves both segmentation accuracy and effect.
引文
[1]李锵,张琦珺,关欣,等.基于改进模糊C均值算法的颈动脉超声图像分割[J].天津大学学报:自然科学与工程技术版,2018,51(1):95-102.Li Qiang,Zhang Qijun,Guan Xin,et al.Segmentation of carotid intima media in ultrasound images using improved fuzzy C means algorithm[J].Journal of Tianjin University:Science and Technology,2018,51(1):95-102(in Chinese).
[2]Yongjian Y,Acton S T.Speckle reducing anisotropic diffusion[J].IEEE Transactions on Image Processing,2002,11(11):1260-1270.
[3]Marsousi M,Plataniotis K N,Stergiopoulos S.Shapebased kidney detection and segmentation in threedimensional abdominal ultrasound images[C]//201436 th Annual International Conference of the IEEE Engineering in Medicine and Biology Society.Chicago,USA,2014:2890-2894.
[4]褚晶辉,王星宇,吕卫.基于帧间相关性的乳腺MRI三维分割[J].天津大学学报:自然科学与工程技术版,2017,50(8):835-842.Chu Jinghui,Wang Xingyu,LüWei.3D segmentation of breast MRI based on inter-frame correlations[J].Journal of Tianjin University:Science and Technology,2017,50(8):835-842(in Chinese).
[5]Li C,Xu C,Gui C,et al.Distance regularized level set evolution and its application to image segmentation[J].IEEE Transactions on Image Processing,2010,19(12):3243-3254.
[6]Yang F,Qin W,Xie Y,et al.A shape-optimized framework for kidney segmentation in ultrasound images using NLTV denoising and DRLSE[J].BioMedical Engineering OnLine,2012,11:82.
[7]于潇.左心室超声心动图降噪和分割方法研究[D].哈尔滨:哈尔滨工业大学电子与信息工程学院,2012.Yu Xiao.Research on Despeckling and Segmentation of Left Ventricular Echocardiography[D].Harbin:School of Electronics and Information Engineering,Harbin Institute of Technology,2012(in Chinese).
[8]Selvathi D,Bama S.Phase based distance regularized level set for the segmentation of ultrasound kidney images[J].Pattern Recognition Letters,2017,86:9-17.
[9]张迎春,郭禾.基于粗糙集和新能量公式的水平集图像分割[J].自动化学报,2015,41(11):1913-1925.Zhang Yingchun,Guo He.Level set image segmentation based on rough set and new energy formula[J].Acta Automatica Sinica,2015,41(11):1913-1925(in Chinese).
[10]何传江,李梦,詹毅.用于图像分割的自适应距离保持水平集演化[J].软件学报,2008,19(12):3161-3169.He Chuanjiang,Li Meng,Zhan Yi.Adaptive distance preserving level set evolution for image segmentation[J].Journal of Software,2008,19(12):3161-3169(in Chinese).
[11]周林,平西建,童莉.一种改进的自适应距离保持水平集演化方法[J].应用科学学报,2011,29(3):274-280.Zhou Lin,Ping Xijian,Tong Li.Improved method of adaptive distance preserving level set evolution[J].Journal of Applied Sciences,2011,29(3):274-280(in Chinese).
[12]翁桂荣,朱云龙,钱森.一种自适应初始轮廓的水平集演化方法的研究[J].电子学报,2017,45(11):2728-2734.Weng Guirong,Zhu Yunlong,Qian Sen.Research on an adaptive level set evolution method for initial contour[J].Acta Electronica Sinica,2017,45(11):2728-2734(in Chinese).
[13]陈雯,朱敏.结合区域间差异性的水平集演化模型[J].华东师范大学学报:自然科学版,2014(6):57-66.Chen Wen,Zhu Min.Level set evolution model with inter-region dissimilarity[J].Journal of East China Normal University:Natural Science,2014(6):57-66(in Chinese).
[14]王斌,李洁,高新波.一种基于边缘与区域信息的先验水平集图像分割方法[J].计算机学报,2012,35(5):1067-1072.Wang Bin,Li Jie,Gao Xinbo.An edge and regionbased level set method with shape priors for image segmentation[J].Chinese Journal of Computers,2012,35(5):1067-1072(in Chinese).
[15]Zhang K,Zhang L,Lam K M,et al.A level set approach to image segmentation with intensity inhomogeneity[J].IEEE Transactions on Cybernetics,2016,46(2):546-554.
[16]Li C,Kao C Y,Gore J C,et al.Minimization of region-scalable fitting energy for image segmentation[J].IEEE Transactions on Image Processing,2008,17(10):1940-1949.
[17]廖祥云,袁志勇,郑奇,等.引入局部全局信息的区域自适应局域化快速活动轮廓模型[J].计算机学报,2016,39(7):1464-1476.Liao Xiangyun,Yuan Zhiyong,Zheng Qi,et al.Adaptive localized region-based fast active contour model by introducing global information in local region[J].Chinese Journal of Computers,2016,39(7):1464-1476(in Chinese).
[18]Liu S,Peng Y.A local region-based chan-vese model for image segmentation[J].Pattern Recognition,2012,45(7):2769-2779.
[19]Wuttisarnwattana P.Automatic whole mouse segmentation for cryo-imaging data using DRLSE model[C]//2016 13 th International Conference on Electrical Engineering/Electronics,Computer,Telecommunications and Information Technology(ECTI-CON).Chiang Mai,Thailand,2016:1-5.
[20]张耀楠,高原,唐亮,等.基于ASM水平集分割方法的心脏同步功能估计[J].中国医疗器械杂志,2016,40(4):240-244.Zhang Yaonan,Gao Yuan,Tang Liang,et al.Cardiac synchronization function estimation based on ASM level set segmentation method[J].Chinese Journal of Medical lnstrumentation,2016,40(4):240-244(in Chinese).
[21]邵昊阳,张英涛,鲜敏,等.基于多域先验的乳腺超声图像协同分割[J].自动化学报,2016,42(4):580-592.Shao Haoyang,Zhang Yingtao,Xian Min,et al.Breast ultrasound image co-segmentation by means of multiple-domain knowledge[J].Acta Automatica Sinica,2016,42(4):580-592(in Chinese).
[22]Torres H R,Queirós S,Morais P,et al.Kidney segmentation in ultrasound,magnetic resonance and computed tomography images:A systematic review[J].Computer Methods and Programs in Biomedicine,2018,157(2018):49-67.
[23]Shan J,Cheng H D,Wang Y.Completely automated segmentation approach for breast ultrasound images using multiple-domain features[J].Ultrasound in Medicine&Biology,2012,38(2):262-275.
[2]Yongjian Y,Acton S T.Speckle reducing anisotropic diffusion[J].IEEE Transactions on Image Processing,2002,11(11):1260-1270.
[3]Marsousi M,Plataniotis K N,Stergiopoulos S.Shapebased kidney detection and segmentation in threedimensional abdominal ultrasound images[C]//201436 th Annual International Conference of the IEEE Engineering in Medicine and Biology Society.Chicago,USA,2014:2890-2894.
[4]褚晶辉,王星宇,吕卫.基于帧间相关性的乳腺MRI三维分割[J].天津大学学报:自然科学与工程技术版,2017,50(8):835-842.Chu Jinghui,Wang Xingyu,LüWei.3D segmentation of breast MRI based on inter-frame correlations[J].Journal of Tianjin University:Science and Technology,2017,50(8):835-842(in Chinese).
[5]Li C,Xu C,Gui C,et al.Distance regularized level set evolution and its application to image segmentation[J].IEEE Transactions on Image Processing,2010,19(12):3243-3254.
[6]Yang F,Qin W,Xie Y,et al.A shape-optimized framework for kidney segmentation in ultrasound images using NLTV denoising and DRLSE[J].BioMedical Engineering OnLine,2012,11:82.
[7]于潇.左心室超声心动图降噪和分割方法研究[D].哈尔滨:哈尔滨工业大学电子与信息工程学院,2012.Yu Xiao.Research on Despeckling and Segmentation of Left Ventricular Echocardiography[D].Harbin:School of Electronics and Information Engineering,Harbin Institute of Technology,2012(in Chinese).
[8]Selvathi D,Bama S.Phase based distance regularized level set for the segmentation of ultrasound kidney images[J].Pattern Recognition Letters,2017,86:9-17.
[9]张迎春,郭禾.基于粗糙集和新能量公式的水平集图像分割[J].自动化学报,2015,41(11):1913-1925.Zhang Yingchun,Guo He.Level set image segmentation based on rough set and new energy formula[J].Acta Automatica Sinica,2015,41(11):1913-1925(in Chinese).
[10]何传江,李梦,詹毅.用于图像分割的自适应距离保持水平集演化[J].软件学报,2008,19(12):3161-3169.He Chuanjiang,Li Meng,Zhan Yi.Adaptive distance preserving level set evolution for image segmentation[J].Journal of Software,2008,19(12):3161-3169(in Chinese).
[11]周林,平西建,童莉.一种改进的自适应距离保持水平集演化方法[J].应用科学学报,2011,29(3):274-280.Zhou Lin,Ping Xijian,Tong Li.Improved method of adaptive distance preserving level set evolution[J].Journal of Applied Sciences,2011,29(3):274-280(in Chinese).
[12]翁桂荣,朱云龙,钱森.一种自适应初始轮廓的水平集演化方法的研究[J].电子学报,2017,45(11):2728-2734.Weng Guirong,Zhu Yunlong,Qian Sen.Research on an adaptive level set evolution method for initial contour[J].Acta Electronica Sinica,2017,45(11):2728-2734(in Chinese).
[13]陈雯,朱敏.结合区域间差异性的水平集演化模型[J].华东师范大学学报:自然科学版,2014(6):57-66.Chen Wen,Zhu Min.Level set evolution model with inter-region dissimilarity[J].Journal of East China Normal University:Natural Science,2014(6):57-66(in Chinese).
[14]王斌,李洁,高新波.一种基于边缘与区域信息的先验水平集图像分割方法[J].计算机学报,2012,35(5):1067-1072.Wang Bin,Li Jie,Gao Xinbo.An edge and regionbased level set method with shape priors for image segmentation[J].Chinese Journal of Computers,2012,35(5):1067-1072(in Chinese).
[15]Zhang K,Zhang L,Lam K M,et al.A level set approach to image segmentation with intensity inhomogeneity[J].IEEE Transactions on Cybernetics,2016,46(2):546-554.
[16]Li C,Kao C Y,Gore J C,et al.Minimization of region-scalable fitting energy for image segmentation[J].IEEE Transactions on Image Processing,2008,17(10):1940-1949.
[17]廖祥云,袁志勇,郑奇,等.引入局部全局信息的区域自适应局域化快速活动轮廓模型[J].计算机学报,2016,39(7):1464-1476.Liao Xiangyun,Yuan Zhiyong,Zheng Qi,et al.Adaptive localized region-based fast active contour model by introducing global information in local region[J].Chinese Journal of Computers,2016,39(7):1464-1476(in Chinese).
[18]Liu S,Peng Y.A local region-based chan-vese model for image segmentation[J].Pattern Recognition,2012,45(7):2769-2779.
[19]Wuttisarnwattana P.Automatic whole mouse segmentation for cryo-imaging data using DRLSE model[C]//2016 13 th International Conference on Electrical Engineering/Electronics,Computer,Telecommunications and Information Technology(ECTI-CON).Chiang Mai,Thailand,2016:1-5.
[20]张耀楠,高原,唐亮,等.基于ASM水平集分割方法的心脏同步功能估计[J].中国医疗器械杂志,2016,40(4):240-244.Zhang Yaonan,Gao Yuan,Tang Liang,et al.Cardiac synchronization function estimation based on ASM level set segmentation method[J].Chinese Journal of Medical lnstrumentation,2016,40(4):240-244(in Chinese).
[21]邵昊阳,张英涛,鲜敏,等.基于多域先验的乳腺超声图像协同分割[J].自动化学报,2016,42(4):580-592.Shao Haoyang,Zhang Yingtao,Xian Min,et al.Breast ultrasound image co-segmentation by means of multiple-domain knowledge[J].Acta Automatica Sinica,2016,42(4):580-592(in Chinese).
[22]Torres H R,Queirós S,Morais P,et al.Kidney segmentation in ultrasound,magnetic resonance and computed tomography images:A systematic review[J].Computer Methods and Programs in Biomedicine,2018,157(2018):49-67.
[23]Shan J,Cheng H D,Wang Y.Completely automated segmentation approach for breast ultrasound images using multiple-domain features[J].Ultrasound in Medicine&Biology,2012,38(2):262-275.