融合注意力机制的阿尔茨海默症识别模型  

Alzheimer's disease recognition model incorporating attention mechanism

作　　者：曾安[1] 高征 ZENG An;GAO Zheng(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院,广州510006

出　　处：《生物医学工程研究》2021年第3期233-240,共8页Journal Of Biomedical Engineering Research

基　　金：国家自然科学基金资助项目(61772143);广东省级科技计划项目(2019A050510041)。

摘　　要：本研究针对阿尔茨海默症(Alzheimer′s disease,AD)不同阶段人群难以识别区别的问题,提出一种融合注意力机制的AD识别模型。该方法利用脑区模板标签(Automated anatomical labeling,AAL)划分受试者的大脑区域,将同一脑区中具有相同属性价值的体素数据组织在一起,并分别构造每个脑区所对应的基分类器。同时,受深度学习与计算机视觉注意力机制相关工作的启发,提出一种直映式注意力机制,提高了识别模型的准确率以及稳定性。通过利用直推式支持向量机对基分类器的设计进行优化,进一步提高了识别模型的准确率。实验结果表明,该方法具有良好的分类效果,为其它脑疾病诊断提供了新思路。Aiming at Alzheimer′s Disease(AD)at different stages are difficult to identify and distinguish,we proposed an AD recognition model that incorporated attention mechanism.Automated anatomical labeling(AAL)was used to divide the brain area of the subject,the voxel datas with the same attribute value in the same brain area were organized together,and the base classifiers corresponding to each brain area were constructed respectively.At the same time,inspired by the work related to the attention mechanism of deep learning and computer vision,a direct mapping attention mechanism was proposed to improve the accuracy and stability of the recognition model.By using the twin support vector machine to optimize the design of the base classifiers,the accuracy of the recognition model was further improved.Experimental results show that this method has a good classification effect,can provide a new idea for the diagnosis of other brain diseases.

关 键 词：阿尔茨海默症 注意力机制 体素 基分类器 深度学习 直推式支持向量机 

分 类 号：R318[医药卫生—生物医学工程]