基于直方图拟合的MRI图像水平集分割  

MR Image Segmentation Based on Level Set and Histogram Analysis

作　　者：陈志彬[1,2] 邱天爽[1] Su Ruan 

机构地区：[1]大连理工大学电子与信息工程学院,大连116024 [2]辽宁科技大学电子与信息工程学院,鞍山114051 [3]法国兰斯大学电气与工业计算机系,Troyes Cedex 10026

出　　处：《中国生物医学工程学报》2008年第1期1-7,共7页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金资助项目(30170259,30570475,60372081);教育部博士点基金资助项目(20050141025)

摘　　要：利用混合高斯模型对MRI图像直方图进行分析,将拟合获得的特征参数作为水平集曲线进化的约束条件,对医学图像进行分割。分割中采用的自适应Level Set方法,能够自适应地确定曲线进化方向(扩张或收缩),而不必在分割之前指定其进化方向,减少了人工干预;同时也克服了传统测地活动轮廓线(GAC)方法对图像梯度信息的过分依赖,以及由于对图像进行大尺度高斯平滑处理造成边缘点移动、定位准确度下降的缺陷。分别对MRI仿真和真实图像进行了实验,MRI仿真实验的分割敏感性、专一性和总体性能指标分别达到了94.72%、97.52%和97.22%。分割结果的定量分析和定性分析表明算法的有效性以及较高的分割准确度。The MRI histogram was analyzed with the Gaussian mixture model, and the image was segmented with the adaptive Level Set method, in which the characteristic parameters obtained from the histogram fitting were used to constraint the curves evolution. The adaptive method is able to determine the direction of the curves evolution adaptively without the manual intervention while disusing the evolution direction pre-determined before the segmentation procedure begin. This new method does not depend excessively on the gradient information as that in the Geodesic Active Contour （GAC） method, and it overcomes the drawback caused by the Gaussian smoothing to the image. The experiments with the artificial images and the real MR images were conducted and the quantitative and qualitative analyses demonstrated the validity and accuracy in image segmentation.

关 键 词：直方图拟合 图像分割 LevelSet方法 K-均值 

分 类 号：R318[医药卫生—生物医学工程]