基于统计学习的多层医学图像语义建模方法  

A multi-level medical image semantic modeling approach based on statistical learning

作　　者：林春漪[1] 尹俊勋[2] 高学[2] 陈建宇[3] 孙少晖[1] 

机构地区：[1]中山大学生物医学工程系,广州510080 [2]华南理工大学电子与信息工程学院,广州510640 [3]中山大学第二附属医院,广州510120

出　　处：《深圳大学学报（理工版）》2007年第2期138-143,共6页Journal of Shenzhen University(Science and Engineering)

基　　金：国家自然科学基金资助项目(60472063)

摘　　要：提出一种在小样本的情况下,基于多层贝叶斯网络的医学图像语义建模方法.该方法采用支持向量机实现从低层视觉特征到对象语义的映射,使用贝叶斯网络融合对象语义,提取高级语义,从而建立一个多层医学图像语义模型,可支持多层次的医学图像语义自动标注及其检索.将该方法用于星形细胞瘤恶性程度的语义提取,并建立一个多层语义模型.实验表明,该模型与使用K近邻分类器或高斯混合模型取代SVM的语义模型相比,查全率有明显的提高.A semantic modeling approach for medical image semantic retrieval based on hierarchical Bayesian networks was proposed, in a small set of samples. SVMs （support vector machines） was used to map low-level image features into object semantics, then high-level semantics was captured through fusing these object semantics using a Bayesian network. A multi-layer medical image semantic model was built to aim to enable automatic image annotation and semantic retrieval by using various keywords at different semantic levels. To validate the method, a multilevel semantic model was built from a small set of astrocytona MRI （ magnetic resonance imaging） samples, in order to extract semantics of astrocytona in malignant degree. Experiment results show that this approach is effective to enable multi-level interpretation of astrocytona MRI. It out performs the Bayesian network-based models using k-nearest neighbor classifiers （K-NN） or Gaussion mixture models （GMM）.

关 键 词：多层贝叶斯网络 支持向量机 语义建模 医学图像 机器学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]