基于瞬时转变率模型的脑网络状态观测算法  

A brain network state observation algorithm based on instantaneous transition rate model

作　　者：刘畅 王彬[1] 薛洁[2] 熊新[1] 郭子洋 LIU Chang;WANG Bin;XUE Jie;XIONG Xin;GUO Zi-yang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500;Faculty of Information Network Security,Yunnan Police Officer Academy,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500 [2]云南警官学院信息网络安全学院,云南昆明650500

出　　处：《计算机工程与科学》2019年第7期1325-1334,共10页Computer Engineering & Science

基　　金：国家自然科学基金(81470084,81771926,61463024,61763022,61263017)

摘　　要：针对K-means等聚类方法在脑网络状态观测中稳定性和鲁棒性较差的缺点,提出了一种基于瞬时转变率模型的脑网络状态观测算法。通过对状态转换临界点进行分组统计和分析,计算每一个临界时间点的状态瞬时转变率,在此基础上构建脑网络状态观测算法,并使用区间估计方法对状态转换的观测效果进行估计和验证。在脑网络数据库样本中的实验结果显示,与K-means等脑网络状态聚类观测算法相比,该算法在不同条件下的聚类稳定性更好,对样本差异的适应性更强,受参数选择的影响更小,能直观地观测到脑网络状态转换趋势。Aiming at the poor stability and robustness in brain network state observation in clustering methods such as K-means,we propose a brain network state observation algorithm based on instantaneous transition rate model.The algorithm calculates the instantaneous state transition rate of each critical time point by group statistics and analysis.Based on this,we construct a brain network state observation model,and estimate and verify the state transition observation effect by the interval estimation method.Experimental results of the brain network database samples show that compared with the K-means and other brain network state clustering observation algorithms,the proposed algorithm has better cluster stability under different conditions and is more adaptable to individual sample differences.It is less affected by parameter selection and can visually observe the trend of brain network state transition.

关 键 词：状态观测 动态功能连接 高维聚类 静息态fMRI 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]