SFECNet: Fusion of Structural, Functional and Effective Connectivity Network for Epilepsy Classification

For brain network analysis, multimodal fusion learning leverages multiple medical imaging modalities to gather more comprehensive information. This method could usually results more accurate diagnoses in brain diseases like epilepsy. Many multimodal approaches typically construct only structural connectivity (SC) networks and functional connectivity (FC) networks. However SC and FC are undirected connectivity, leading to lack of directed interaction between brain regions. Effective connectivity (EC) networks implying causal relationships, could further offer complementary information. So we propose a Fusion of Structural, Functional and Effective Connectivity Network for Epilepsy Classification (SFECNet). We construct SC, FC, and EC networks using DTI and fMRI data and apply graph convolutional networks (GCNs) to extract dynamic and static features. Crossattention mechanisms guide the fusion of multimodal feature information to finally make epilepsy classification. Experiments on the epilepsy dataset demonstrate the feasibility and effectiveness of our proposed method.