阿尔茨海默病的图神经网络分类方法研究进展

Research Progress on Graph Neural Network Classification Methods for Alzheimer′s Disease

阿尔茨海默病(AD)是一种不可逆的神经退行性疾病,会导致认知能力的逐渐下降。AD症状的演变过程可能很长,在不同的神经成像模式中可检测到脑区生物标志物的细微变化,但其早期检测具有挑战性。由于神经成像数据的高度复杂性和大脑网络的不规则性,传统的机器学习和深度神经网络模型存在许多不足,开发基于图神经网络(GNN)的计算机辅助诊断(CAD)模型可以为分析非欧几里得空间的神经影像模式以及探究生物标志物提供极大帮助。首先,对基于GNN分类方法的AD预测进行详细的调研和概述。然后,从基于单模态数据和基于多模态数据两个视角进行梳理,重点介绍和分析这些方法在单模态和多模态数据应用场景中的数据提取、脑网络建模、特征学习、信息融合等过程,并评述部分方法的性能。最后,针对GNN应用于AD诊断的主要挑战和未来研究方向进行了展望,为AD辅助诊断的进一步研究提供有益的建议。

Alzheimer′s Disease(AD)is an irreversible neurodegenerative disorder that leads to gradual cognitive decline.The evolution of AD symptoms can be long,with subtle changes in biomarkers in brain regions that are detectable by different neuroimaging modalities;however,early detection is challenging.Given the high complexity of neuroimaging data and the irregularity of brain networks,traditional machine learning,and deep neural network models exhibit many shortcomings,and the development of Computer-Aided Diagnostic(CAD)models based on Graph Neural Network(GNN)can be beneficial for probing biomarkers and analyzing neuroimaging patterns in non-Euclidean space.First,a detailed investigation and overview of AD prediction based on GNN classification methods is carried out.Subsequently,an analysis is conducted from the two perspectives of single-and multi-modal data,with a focus on discussing and analyzing the processes of data extraction,brain network modeling,feature learning,and information fusion within the context of single-and multi-modal data applications.A performance evaluation is provided for certain methods.Finally,the primary challenges and future research directions for the application of GNNs in AD diagnosis are outlined to provide beneficial suggestions for further research on AD-assisted diagnosis.