基于遗传密度峰值聚类的医学图像分割

摘要

To get the optimal segmentation threshold of density peaks clustering (DPC) algorithm and satisfactory segmentation effect, a method based on genetic algorithm to obtain the maximum entropy of the segmenting image was proposed.For some defects in DPC algorithm, such that cut-off distance dcis given using DPC algorithm relied on prior knowledge, subjective randomness in cluster centers was selected by manual work.The proposed algorithm actualized the adaptive segmentation of DPC algorithm and it was applied to medical image.Simulation experiments adopted classic disease images from the whole brain atlas website of Harvard medical school and results were compared with that of K-means, affinity propagation clustering algorithm and DPC algorithm.These experiments show that the improved algorithm of DPC can automatically obtain cut-off distance and confirm the cluster centers, and it gets better segmentation results.%针对密度峰值算法 (density peaks cluster, DPC) 依靠先验知识给定截断距离dc且人工选择聚类中心点具有主观随意性等缺陷, 提出一种基于遗传算法求取分割图像最大熵值, 获得最优分割阈值的方法.得到满意的分割效果, 实现了DPC算法的自适应分割并应用到医学图像上.仿真实验采用多张哈佛全脑图中的经典疾病图像, 与K-means、AP (仿射传播) 聚类算法及DPC算法作比较, 比较结果表明, DPC的改进算法能自动获取截断距离, 确定聚类中心, 获得更好的分割效果.