【摘要】 阿尔茨海默病是一种严重影响人类生活的病症,它具有难以治愈的特点.而其早期症状,轻度认知障碍的诊断就成了延缓发展和治疗的关键.核磁共振图像是诊断脑部疾病的重要影像资料.通过分析核磁共振图像,再利用分类算法,将轻度认知障碍患者从正常人中区分开来成为一种重要的方法.而特征选择则是提高分类准确率的必要步骤.本文提出将互信息和皮尔逊相关系数集成的特征选择方法,不仅考察每个特征对类标签的相关性,而且保证选出的特征子集之间冗余度最小.实验结果证明,与互信息和mRMR方法结合支持向量机进行分类性能比较,本文提出的方法分类准确性更高,说明本文的特征选择方法具有较好的优势.更多还原

【Abstract】 Alzheimer is a disease which effects our lives.It is difficult to cure.However,the diagnose of mild cognitive impairment,the early stage of Alzheimer,is the key to delay the progress and treatment of the diseas.MRI is a kind of important image data.Analysis of MRI,use of classification algorithm and separating MCI from normal control is a significant method.And feature selection is an essential step to improve the accuracy of classification.We proposed integrated feature selection method combining the mutual information and Pearson correlation coefficient,not only investigating the correlation between each feature and class labels,and ensuring minimum redundancy between the selected feature subsets.Compared with the classification model of support vector institutions with single mutual information method and max- relevance and min- redundancy method,the results show that the proposed method of classification in higher prediction accuracy,illustrating certain advantages.更多还原