遗传K-均值初始化的t混合模型医学图像聚类*

针对基于混合模型的图像聚类质量易受混合模型参数初始值的影响，提出一种遗传K-均值初始化的t混合模型医学图像聚类方法。该方法构建一种医学图像的t混合模型，将遗传算法与K-均值算法相结合，实现对医学图像的初始划分，从而获得混合模型的初始参数，有效克服混合模型对参数初始值选择的敏感性问题，用EM算法多次迭代估计t混合模型参数；最后根据得到的混合模型对医学图像进行聚类。实验表明，该方法实现了医学图像较精准的聚类，有较好的稳定性和通用性。

t-mixture model clustering based on genetic K-means initialization for medical images

The clustering quality of the mixture model clustering for images is vulnerable to the initial values of the mixture model parameters. To solve this problem, this paper proposed one method that based on initialization of genetic K-means algorithm of t mixture model for medical images. It built a t mixture model of medical image, and integrated genetic algorithm with K-means algorithm to realize the initial division of medical images, and then got the initial values of the mixture model. It could effectively overcome the sensitivity of mixture model to the initial selected parameter. Used EM algorithm to estimate the parameters of t mixture model. Finally, clustered the medical images at the base of the proposed mixture model. Experimental results show that medical images can be clustered accurately and the algorithm has great versatility and robustness.