基于RBM-KNN的脑部磁共振图像分类

摘要

The Restricted Boltzmann machine (RBM) and K-nearest neighbor (KNN) classifier method are proposed in this paper,in order to accelerate the classification speed and improve the classification accuracy of medical image.Firstly,a visible binary and hidden binary RBM are constructed,then the RBM is used for training a feature extractor,the extractor can reduce the feature dimensions at the same time.Thirdly,image features are extracted by the feature extractor from the pixel unit directly.Finally,the features are classified by KNN,and the classification accuracy is exampled by test samples.The method is used in the classification of brain magnetic resonance image database,and experimental results show that method presented in the paper show better classification accuracy and outperforms other methods based on single feature extraction in the aspect of classification accuracy rate.%为加快医学图像分类速度,提高分类精确率,文中采用受限玻尔兹曼机(Restricted Bo-ltzmann Machine,RBM)结合K近邻(K-Nearest Neighbor,KNN)分类器方法.首先构建可视层二值对隐层二值RBM,利用RBM训练得到特征提取器,该特征提取器可同时实现特征降维,然后特征提取器从像素单元直接提取图像特征,最后用KNN将特征分类,并用测试样本检验分类准确性.将文中方法用在脑部磁共振图像数据库分类中,实验结果表明,提出的方法具有良好的分类准确率,且明显高于基于单一统计特征提取的医学图像分类方法.