Graph neural network based heterogeneous propagation scheme for classifying Alzheimer's disease

Abstract

Deep learning frameworks for disease classification with neuroimaging and non-imaging information require the capability of capturing associative information among subjects as well as individual features. Graphs represent interactions among nodes, which contain individuals’ feature, through edges in order to incorporate inter-relatedness among heterogeneous data. Previous graph-based approaches for disease classification have focused on the similarities among subjects by establishing customized functions or solely depended on imaging features. The purpose of this paper is to propose a novel graph-based deep learning architecture for classifying Alzheimer’s disease (AD) by combining resting state functional MRI and demographic measures without defining any study specific function. We used neuroimaging data from ADNI and OASIS database to test the robustness of our proposed model. We combined individuals’ imaging-based features and non-imaging information by categorizing into distinctive nodes to construct subject-demographic bipartite graph. The approximation personalized propagation of neural predictions (APPNP), a recently developed graph neural network model, is used to classify the AD continuum from cognitive unimpaired individuals. The results showed that our model successfully captures the heterogeneous relations among subjects and improved the quality of the classifications compare to other classical and deep learning models and our model outperformed among them.