基于小波域隐马尔可夫模型的人脑MRI体数据分类  

Classification of Brain MRI Volume Data Based on Wavelet-Domain Hidden Markov Model

作　　者：刘正光[1] 孙嘉[1] 林雪燕[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《天津大学学报》2007年第1期116-120,共5页Journal of Tianjin University(Science and Technology)

摘　　要：为保证三维体视化图像能较准确地表达组织,以人脑磁共振图像为例,提出了基于小波域隐马尔科夫模型的体数据分类算法,首先采用EM算法进行HMT模型参数估计,然后通过小波分解,得到近似初始分类数和各类在小波空间中的特征量,这在以往体数据分类中需要事先对体数据进行大量的训练才能得到．分类结果采用ICM (iterated conditional mode)方法获得．其结果表明,该方法在运算时间和分类效果上都优于以往的多分辨率分类方法．To ensure the 3D visualization to render the tissues accurately, a new classification of brain magnetism resonance imaging volume data was proposed based on wavelet-domain hidden Markov model. First, parameters of HMT were estimated by expectation-maximization （EM） algorithm. Then the approximative class and the eigenvalue of each class in wavelet space, which could be obtained only by lots of training using the existent algorithms, were gained by proper wavelet-decomposition. The result of classification was obtained by iterated conditional mode algorithm, which shows that the proposed algorithm is more accurate but less complex.

关 键 词：小波 隐马尔可夫模型(HMM) 磁共振图像(MRI) 最大期望(EM)算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]