基于高斯混合密度模型的医学图像聚类方法

研究了医学图像的聚类问题,提出一种基于高斯混合密度模型的K-EM聚类算法,并将此算法用于人体腹部图像数据,实现肝、肾、脾等主要器官的分类.在算法中,随机选取腹部图像像素数据,用QAIC信息准则确定训练样本的最佳类别数;用K均值聚类算法得到混合模型的初始参数;用期望最大(EM)算法多次迭代建立腹部图像数据的混合密度模型;运用贝叶斯准则,将腹部图像所有像素值划分到混合模型中相应的模型分支,得到每个器官像素值划分的正确率与误判率.试验结果表明,新算法分类的平均正确率高于85%、误判率低于10%,优于K均值算法.

Clustering of medical image based on Gaussian mixture density model

Problem of medical image clustering was studied.A kind of K-EM clustering algorithm based on Gaussian mixture density model was proposed.Human abdominal image data was processed with this algorithm,in order to classify liver,kidney,spleen and other major organs.In this algorithm,abdominal image pixel data was randomly selected,the QAIC information criterion was used to determine the best class number of training samples,and K-means clustering algorithm was used to get initial parameters of this mixture mode...