改进的变分水平集算法在灰度不均匀MRI图像分割中应用(自动保存的)(修复的).docVIP

改进的变分水平集算法在非均质MRI图像分割中应用* 王昌 秦鑫 于毅 新乡医学院 生物医学工程学院 （新乡 453003） 对于灰度不均匀的MRI图像，普通C-V模型不能进行有效分割。为此本文提出了一种改进的变分水平集算法，在对不均匀组织进行分割同时用偏置场原始图像。首先依据图像灰度不均匀效应模型，定义了基于局部灰度的K均值聚类准则，并将聚类准则项整合为变分水平集函数的基本能量项，构造变分水平集的能量函数，求解能量函数的欧拉-拉格朗日方程。将本算法的分割结果与CV模型及LBF模型的分割结果进行比较，该算法可在分割灰度不均匀MRI图像的同时利用偏置场对灰度不均匀场进行校正，有很好的临床应用价值。 关键词：变分水平集算法、灰度不均匀、偏置场、K均值聚类MRI图像分割 The application of improved variational?level set?in?the?segmentation of MRI image with intensity inhomogeneity Wang Chang Qin Xin Yu Yi School of Biomedical Engineering, Xinxiang Medical University, Xinxiang 453003, China Abstract: It is difficult to segment the MRI images with intensity inhomogeneity by C-V active contour model. So the improved variational level set formulation was proposed in this paper to segment tissue with the intensity inhomogeneity and simultaneously estimate bias field to correct the source image. Firstly based on the image model with intensity inhomogeneity, the local K-means clustering criterion was defined. Then the clustering criterion function was integrated into the basic energy term of variational level set formulation, and the energy function of the variational level set formulation was constructed. The energy function was minimized by solving Euler-Lagrange equation. By comparing with the segmentation results of C-V model and LBF model, this algorithm can segment the MRI image with intensity inhomogeneity and estimate bias field to correct the source image. Our algorithm has good clinical application value. Key words: variational level set, intensity inhomogeneity, bias field, K-means clustering, MRI image segmentation 1 引言 当前成像过程中由于造成图像具有灰度不均匀的特点[1]，影响影像医生对病变区域识别和判断，不同医生对目标和病变区域识别也不完全一致。如何准确分割灰度不均匀的图像成为当前研究热点。基于能量泛函的水平集分割算法可利用不断演化的闭合曲线来获取非闭合的目标和轮廓，水平集算法是当前分割提取目标的主流算法。水平集算法包括边缘型和区域型两类。Caselles V提出Geodesic Active Contours) [2]，该模型是基于边缘型的水平集依赖图像局部的边缘信息，可用于强边缘目标的分割，但是对边缘变化、无明显梯度变化的弱边缘分割效果差，并且容易受到噪声的影响。Cha和Vese提出Chan-Vese(C-V模型)[3]是对Mumford-Shah能量泛函优化，水平集的演化，可以分割无边缘模糊、无梯度意义的图像，且对噪声不敏感，但不能分割灰度不均匀的图像