Brain-Inspired Fuzzy Graph Convolution Network for Alzheimer's Disease Diagnosis Based on Imaging Genetics Data

Abstract

The analysis of multiomics biomedical data has become increasingly critical in clinical decision-making for brain diseases, such as Alzheimer's disease (AD). However, the inherent fuzziness of biomedical information limits the classification performance of existing methods, and current disease models struggle to explore pathogenetic mechanisms. Facing with these issues, this article develops a fuzzy graph-based deep learning method to achieve accurate diagnosis and pathogeny identification for brain diseases. First, fuzzy graphs are constructed to describe the associations between pathogenies using fuzzy memberships. Second, a mathematical model inspired by the fuzzy mechanisms of brain is established, effectively capturing the fuzzy congregation patterns of feature information across brain regions and genes. Finally, a brain-inspired fuzzy graph convolutional network (BI-FGCN) is proposed. In BI-FGCN, white-boxed convolutional operations are designed based on the mathematical model. Experimental results across multiple brain disease datasets demonstrate the superiority of BI-FGCN in AD diagnosis and pathogeny identification. We provide a reliable supporting method for the diagnosis and treatment of brain diseases.