基于多模态影像特征的AD诊断研究  

Research on Alzheimer's disease diagnosis using multi-modal image features

作　　者：张格 林岚[1] 康文杰 吴水才[1] ZHANG Ge;LIN Lan;KANG Wen-jie;WU Shui-cai(Intelligent Physiological Measurement and Clinical Translation,Beijing International Base for Scientific and Technological Cooperation,Department of Biomedical Engineering,Faculty of Environment and Life,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学环境与生命学部生物医学工程系,智能化生理测量与临床转化北京市国际科研合作基地,北京100124

出　　处：《医疗卫生装备》2021年第12期1-6,16,共7页Chinese Medical Equipment Journal

基　　金：国家自然科学基金项目(81971683);北京市自然科学基金-海淀原始创新联合基金项目(L182010)。

摘　　要：目的:探究多模态融合特征对阿尔茨海默病(Alzheimer’s disease,AD)病程诊断性能的影响。方法:以阿尔茨海默病神经影像学计划数据库中81例受试者的多模态影像数据为研究对象,在卷积神经网络提取的T_(1)加权成像(T_(1) weight image,T_(1)WI)特征基础上,利用图卷积神经网络(graph convolutional neural networks,GCN)提取弥散张量成像(diffusion tensor imaging,DTI)脑白质网络特征,然后采用类别提升集成算法融合多模态特征进行AD诊断预测。结果:在DTI脑白质网络特征学习中,GCN模型的诊断准确率为80.0%,优于传统机器学习模型;在单、多模态的AD诊断性能比较中,基于多模态特征的诊断准确率为85.3%,优于T_(1)WI单模态特征。结论:各模态特征间存在一定互补性,GCN可以提取DTI影像中更具表征性的脑连接网络特征,与T_(1)WI特征融合可进一步提高AD的诊断性能。Objective To explore the effect of multi-modal fusion features on the diagnosis performance of Alzheimer's disease(AD).Methods The multi-modal imaging data of 81 subjects were selected from the Alzheimer's Disease Neuroimaging Initjative database,the T_(1) weight image(T_(1)WI)features were extracted by the convolutional neural network,a graph convolutional neural networks(GCN)was used to obtain the features of the brain white matter network in diffusion tensor imaging(DTI)images,and then categorical boosting(CatBoost)ensemble algorithm was used to fuse multi-modal features for AD diagnosis.Results In the DTI brain white matter network feature learning,the diagnosis accuracy of the GCN model was 80.0%,which was better than the traditional machine learning model;the diagnosis accuracy based on multi-modal features was 85.3%,which was higher than that based on the T_(1)WI unimodal features.Conclusion There is a certain complementarity among the modal features,GCN can extract representational features of brain connectivity network from DTI images,and fusion with T_(1)WI features can further improve AD diagnosis performance.[Chinese Medical Equipment Journal,2021,42(12):1-6,16]

关 键 词：多模态影像特征 特征融合 图卷积神经网络 脑连接网络 AD诊断 

分 类 号：R318[医药卫生—生物医学工程] R445.2[医药卫生—基础医学]