Evo-TGSF: An Evolutionary Topology-Genomic Subspace Fusion Framework for Alzheimer’s Disease Classification using Imaging-Genomics Data

Brain imaging genomics offers a powerful paradigm for unraveling the complex pathophysiological mechanisms underlying Alzheimer’s disease (AD). However, effective integration of high-dimensional functional magnetic resonance imaging (fMRI) data and single nucleotide polymorphism (SNP) data remains challenging. In this paper, we propose an evolutionary topology-genomic subspace fusion framework, termed Evo-TGSF, for three-class and four-class Alzheimer’s disease classification using fMRI and SNP data. First, we design a topology-augmented graph convolutional network (TA-GCN) for the fMRI modality, in which node representations are initialized using nine graph-theoretic measures derived from the brain connectivity toolbox. Second, we introduce a genomic subspace interaction network (GSI-Net) for the genomics modality, which employs a subspace tokenization strategy to project unordered SNP profiles into latent functional tokens, enabling the modeling of macroscopic epistatic interactions through self-attention mechanisms. Subsequently, we design an adaptive cross-modal gated fusion (ACM-GF) module, achieving the effective synthesis of heterogeneous features. Finally, to optimize the parameters, we propose an improved cuckoo–catfish optimizer (ICCO). Experimental results on the ADNI dataset demonstrate that Evo-TGSF framework achieves classification accuracies of 82.1%, 92.2%, 83.4%, 88.7% and 80.2% for HC/EMCI/LMCI, HC/EMCI/AD, EMCI/LMCI/AD, HC/LMCI/AD and HC/EMCI/LMCI/AD, respectively.