摘要
目前,多模态神经影像数据被广泛应用于阿尔茨海默症(AD)分类研究,然而,这些研究方法存在着一定缺陷。首先,原始特征维度较高且包含大量冗余信息,需要大量的计算资源进行处理和分析。其次,同一患者的完整多模态数据不易获得,导致样本数量较少,不利于训练出可靠的分类模型。为此,提出了一种基于最大相关-最小冗余和典型相关性分析的AD分类方法(mRMR-CCA)。首先使用最大相关-最小冗余(mRMR)算法从各模态数据选择出与分类标签最相关的判别性特征,然后使用典型相关性分析(CCA)算法来挖掘多模态数据间的内在联系,最后进行分类。使用阿尔茨海默氏病神经影像学倡议数据库中的数据进行了实验,实验结果证明了此方法在AD分类中的有效性。
At present,multi-modal neuroimaging data is widely used in Alzheimer s disease(AD)classification research.However,these research methods have certain limitations.Firstly,the original feature dimension is high,and it contains a large amount of redundant information,requiring a large amount of computational resources for processing and analysing.Secondly,it is difficult to obtain complete multi-modal data from the same patient,resulting in a small sample size,which is not conducive to training reliable classification models.To address these issues,this paper proposes an AD classification method based on maximum relevance-minimum redundancy(mRMR)and canonical correlation analysis(CCA).Specifically,the mRMR algorithm is first used to select discriminative features that are most relevant to the classification label from each modality data.Then,the CCA algorithm is used to explore the inherent connections among multi-modal data,and finally,classification is performed.This paper conducted experiments by using data from the Alzheimer s Disease Neuroimaging Initiative database,and the results demonstrated the effectiveness of the proposed method in AD classification.
作者
张国栋
董爱美
齐志云
刘思迪
ZHANG Guodong;DONG Aimei;QI Zhiyun;LIU Sidi(Faculty of Computer Science and Technology,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)
出处
《齐鲁工业大学学报》
CAS
2023年第6期24-33,共10页
Journal of Qilu University of Technology
基金
山东省自然科学基金(ZR2022MF237,ZR2020MF041)。
关键词
阿尔茨海默症
多模态数据
模态融合
特征选择
Alzheimer s disease
multi-modality data
modality fusion
feature selection
作者简介
张国栋,硕士生,研究方向:计算机技术;通信作者:董爱美,博士、副教授,研究方向:机器学习与模式识别,amdong@qlu.edu.cn。
