摘要
目前医学图像数量巨大,利用计算机处理医学图像从而辅助医疗诊断是医学领域研究的热点。根据脑部图像具有对称性的特点,选择支持向量机-递归特征消除(SVM-RFE)算法对融合特征进行特征选择过程中,引入Pearson系数衡量特征信息的冗余度,将特征相关性指标融入SVM-RFE特征子集的筛选标准中,提升了融合特征的分类性能。在一级分类基础上,基于特征学习方法,构建了2Layer-RBM-KNN二级脑部图像分类模型,增加网络深度以进行更高层次的特征抽象,并且结合数据集探究了分类器的选择,实现样本再分类。
Owing to massive medical images and powerful computer processing capability,computer-aided diagnosis works in analysing medical images and becomes a research hotspot currently in the cross field of computer science and medicine.According to the symmetry of brain images,in the process of feature selection for fusion by SVM-RFE algorithm,the pearson coefficient,used to measure the feature redundancy,is integrated into the screening criteria of the SVM-RFE feature subset.Thus,the classification performance of the fusion feature is improved.A multi-level classification model for brain image is constructed.Based on the approximate symmetry characteristics of the brain image,the first-level classification model for brain images is built,and a gray cosine similarity classification method was proposed to classify brain images initially.After that,a method based on feature learning is put forward named 2 Layer-RBM-KNN,it increases the network depth for a greater feature abstraction,explores the classifiers choice with dataset,achieving sample reclassification.
作者
刘承裕
Liu Chengyu(Guilin People's Hospital of Pingle County,Guilin 542400,China)
出处
《国外电子测量技术》
2020年第11期28-33,共6页
Foreign Electronic Measurement Technology
关键词
脑部图像识别分类
特征融合
RBM
KNN
brain image recognition and classification
feature fusion
RBM
KNN
