Constructing brain functional networks with adaptive manifold regularization for early mild cognitive impairment

Brain functional network (BFN) has emerged as a practical path to explore biomarkers for early mild cognitive impairment (eMCI). Currently, most of BFNs only considered the topology structure between two brain regions and ignored the high-order information among multiple brain regions. We proposed an adaptive manifold regularization method to construct a new BFN. Firstly, a traditional hypergraph was constructed through a low-order BFN. Then, an adaptive hypergraph was obtained by updating the traditional hypergraph weight and structure through adaptive hypergraph learning. An adaptive hypergraph manifold regularization term was constructed by the Laplacian matrix of the adaptive hypergraph. Finally, the low-order BFN was optimized through the adaptive hypergraph manifold regularization and L1$$ {L}_1 $$ sparse regularization. The experimental results confirmed that the proposed method outperformed other state-of-the-art methods in classification performance and stability. This study revealed the causes of changes in topological properties and provided a reference for the clinical diagnosis of eMCI.