基于近似密度初始化的医学图像混合模型聚类

The Gaussian Mixture Model Clustering of Medical Image Based on Initialization of Approximate Density Function

基于EM(ExpectationMaximization)的混合模型聚类的效果与参数的初始值存在密切的关系.提出了一种基于近似密度的EM参数初始化方法,该方法用近似密度估计聚类样本点,再根据每个聚类统计EM的混合比、均值、协方差参数的初始值.并应用于人体腹部医学图像数据的高斯混合模型聚类分析,实验结果表明该方法比Kmeans随机初始化方法有更好的聚类效果.

The performance of EM algorithm heavily depends on the initial values of the parameters in EM. In this paper, The approximate density function is adopted to initialize EM. The method estimate samples by the approximate density function and statistics mixturerate, mean value, and covariance. The application of these parameters in analysis of Gaussian Mixture Desity Mode based on real human abdomen medical images and the results of experiments show that it can achieve better effect than Kmeans and random initialization.