摘要
功能性磁共振成像(FMRI)是通过探索神经元活动的生物特性来揭示大脑功能的空间和时间信息的神经影像技术,自问世以来迅速应用于医学和脑神经科学的研究。本文针对目前大脑FMRI图像分割方法必须事先给定分割数目的问题,将完备数据似然方法、分层先验和可识别先验引入混合模型,建立一种对先验弱依赖的贝叶斯分层混合模型,采用可逆跳马尔可夫链蒙特卡罗算法对模型参数进行估计,实现在可变维参数空间的跳跃式抽样。对人类大脑FMRI图像的实例分析中,抽样得到的组织成分数与图像强度特征相一致,抽样分割结果图与原始图像相吻合。数据分析结果说明,贝叶斯分层混合模型方法能较好地体现FMRI图像的实际特征。
Functional Magnetic Resonance Imaging（FMRI） is a neuroimaging technique that exploits the biological demands of neuronal activity to provide spatial and temporal information on brain function.It has been rapidly applied in clinical medicine and brain neuro science.In this paper,a Bayesian hierarchical mixture model,which depends less on priors,is constructed in order to confirm the number of components by data driven approach.Integrity data likelihood method,hierarchical and recognizable prior are brought into the model.Reversible jump Markov chain Monte Carlo（RJMCMC） method is used to sample from changeable dimension parameter spaces.This specific sampling techinique makes the estimation of model parameters more reasonable and practical.It is found from the analysis of brain FMRI data that the estimated number of components is consistent with the image intensity,the segmented image is almost the same as the original.Simulation result shows that Bayesian hierarchical mixture model in some sense can successfully deal with human brain FMRI segmentation with unknown number of components.
出处
《数理统计与管理》
CSSCI
北大核心
2015年第4期603-611,共9页
Journal of Applied Statistics and Management
基金
国家自然科学基金项目(11171117)
广东省自然科学基金项目(S2011010002371)
