基于多超图融合的超图神经网络模型构建及阿尔茨海默病分类

Hypergraph Neural Network Model Construction and Alzheimer's Disease Classification Based on Multi-hypergraph Fusion

针对目前超图神经网络构建方法单一化,导致被试特征间的交互信息无法表征,从而影响超图神经网络模型分类性能的问题。提出一种多超图融合技术,融合多个超图为一个超图,从而互补多个超图各自所表征的高阶特征,以此来提高超图神经网络模型的分类性能。具体来说,基于结构磁共振成像数据,使用基于稀疏表示的最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)方法,稀疏组LASSO方法以及覆盖组LASSO方法进行超图构建,然后分别基于超图融合技术将三个单一超图进行融合。接着基于融合的超图,构建超图神经网络模型,最终用于阿尔兹海默症及轻度认知障碍的分类。实验结果表明,本文所提方法的分类准确率达到79.21%,证明了该方法在阿尔兹海默症及轻度认知障碍的分类有较高的准确性和泛化性。

The existing construction method of hypergraph neural network is simple,resulted in the interaction information between the features of the subjects can not be represented,affected the classification performance of the hypergraph neural network model.A hypergraph fusion technology that fuses multiple hypergraphs into one was proposed,complementing the respective characteristics of multiple hypergraphs.The higher-order features represented were used to improve the classification performance of the hypergraph neural network model.Specifically,the hypergraph fusion technology used the least absolute shrinkage and selection operator(LASSO),sparse group LASSO,and overlapping group LASSO based on sparse representation to construct hypergraphs via structure magnetic resonance imaging(sMRI)data and fused three single hypergraphs based on the hypergraph fusion techniques.And then,a hypergraph neural network model was constructed based on the fused hypergraph.Finally,the classification for Alzheimer s disease and mild cognitive impairment was carried out.The results show that the classification accuracy of the proposed method is 79.21%,which demonstrated experimentally the practical performance and generalization ability of the proposed method in Alzheimer s disease and mild cognitive impairment classification.