改进的K-均值聚类算法及其在脑组织分割中的应用

目的：鉴于K-均值聚类算法易受初始聚类中心的影响，初始聚类中心不仅影响聚类速度。还可能使算法陷入局部极小值，得到错误的聚类结果，基于SOM神经网络，提出了一种改进的K．均值聚类算法并将其应用于脑实质分割。方法：首先，由SOM神经网络对图像进行初始聚类，得到＆个聚类中心值；然后，以SOM神经网络获得的k个聚类中心值作为K_均值聚类算法的初始聚类中心对图像进行％．均值聚类，最终获得图像的聚类分割结果。结果：基于SOM神经网络的K-means聚类算法的分割精度为O．9274，K-means聚类算法的分割精度为0．8649。结论：利用改进的K-均值聚类算法对磁共振脑部图像进行了分割实验，结果表明该算法有效改善了K-means聚类算法初始聚类中心选取的盲目性，使聚类结果更为准确、稳定，取得了比单一方法更好的分割结果。

An Improved K-means Algorithm For Brain Tissue Segmentation

Objective： K-means clustering algorithm is easily impacted by initial value. Initial value selection will affect the speed of convergence. Improper initial value may get the algorithm trapped in local minima, and get the wrong clustering results. This paper proposed a K-means clustering algorithm based on SOM neural network which is used for the automatic segmentation of brain tissues （white matter, gray matter and CSF）. Methods： The algorithm first use SOM network to obtain better initial cluster- ing centers, and then takes the results as the initial value of K-means clustering algorithm for clustering. Results： Then this paper used the improved K-means for the segmentation of brain tissues. The accuracy of K-means clustering algorithm based on SOM neural network is 0.9274, while the accuracy of K-means clustering algorithm is 0.8649. Conclusions： The experimental results show that the algorithm has made an effective segmentation of brain tissues in MRI.