重庆理工大学学报2025，Vol.39Issue(11)：117-124,8.DOI:10.3969/j.issn.1674-8425(z).2025.06.014
增强邻接矩阵驱动GNN的AD诊断方法研究
Research on AD diagnosis method using enhanced adjacency matrix-driven GNN
摘要
Abstract
To address the issue of sparsity in Graph Neural Networks(GNN)adjacency matrices,this paper builds a node similarity-based adjacency matrix.Based on this,an Alzheimer's disease(AD)diagnosis model using enhanced adjacency matrix-driven GNN is proposed.First,one sample from the unknown class in the dataset is selected as the classification sample.Then,10 samples are randomly chosen from each known class to construct a graph.Next,an adjacency matrix is built with node similarity based on the different modalities of the node embeddings and adding the adjacency matrix to the adjacency operator family.In each layer of the GNN,the node embeddings are updated using the operators from the adjacency operator family.After updating the node embeddings in the final layer,the classification result is obtained through softmax.Results show this model achieves an F1 score of 0.958 and an accuracy of 96.09％on the Alzheimer's Disease Neuroimaging Initiative dataset,improving the accuracy by 1.99％compared to SOTA models.关键词
图神经网络/阿尔茨海默病/邻接矩阵/节点相似度Key words
GNN/Alzheimer's disease/adjacency matrix/node similarity分类
信息技术与安全科学引用本文复制引用
李修军,赵殿飞,葛雄心,杨菁菁,张昱..增强邻接矩阵驱动GNN的AD诊断方法研究[J].重庆理工大学学报,2025,39(11):117-124,8.基金项目
吉林省教育厅科学技术研究项目(JJKH20220780KJ) （JJKH20220780KJ）
吉林省教育厅项目(JJKH20230847KJ) （JJKH20230847KJ）
