基于图的特征选择算法在阿兹海默症诊断问题研究

发布时间：2018-05-13 22:18

本文选题：阿兹海默病诊断 + 特征选择　；参考：《计算机应用研究》2017年04期

【摘要】：针对神经性疾病难以确诊的问题,提出了一种基于图的特征选择方法,过滤掉不相干的特征,从而方便并且准确地对疾病患者进行诊断。算法首先基于先验知识定义了两种基本关系(特征关系和样本关系);然后将这两种关系嵌入到一个由最小二次损失函数和l2-范数正则化因子组成的多任务学习框架中进行特征选择;最后,将约简得到的降维矩阵送入支持向量机(SVM)中对阿兹海默症患者进行确诊。通过对Alzheimer’s disease neuroimaging initiative(ANDI)的研究数据集进行实验得知,提出算法的分类效果均优于一般常用分类算法,如K最近邻法(KNN)、支持向量机(SVM)等。提出的算法通过考虑特征选择和引入两种数据的内在关系,有效提高了阿兹海默疾病诊断的正确率。
[Abstract]:In view of the difficult diagnosis of neural diseases, a feature selection method based on graph is proposed to filter out uncoherent features, so that the disease patients are easily and accurately diagnosed. First, two basic relationships (feature relations and sample relationships) are defined based on prior knowledge, and the two relationships are embedded in one Feature selection is performed in a multi task learning framework consisting of the minimum two loss function and the l2- norm regularization factor. Finally, the reduced dimension matrix of the reduction is sent to the support vector machine (SVM) for the diagnosis of Alzheimer's patients. Experiments on the data set of the Alzheimer 's disease neuroimaging initiative (ANDI) are carried out. The results of the proposed algorithm are better than the common common classification algorithms, such as K nearest neighbor (KNN) and support vector machine (SVM). The proposed algorithm can effectively improve the accuracy of Alzheimer's disease diagnosis by considering the characteristics selection and introducing the intrinsic relationship between the two kinds of data.

【作者单位】：广西大学计算机与电子信息学院;广西师范大学计算机科学与信息工程学院;
【基金】：国家“973”计划资助项目(2013CB329404) 中国博士后科学基金资助项目(2015M570837) 广西自然科学基金资助项目(2012GXNSFGA060004,2015GXNSFCB139011) 国家自然科学基金资助项目(61450001,61263035,61573270)
【分类号】：R749.16;TP18

【相似文献】