A lesion-prior guided hyperedge optimization algorithm for early Alzheimer's disease prediction

Alzheimer's disease (AD) is a neurodegenerative disorder characterized by insidious onset and progressive deterioration, with distinct patterns of regional brain impairment and varying sequences of functional decline during disease progression. Hypergraphs can effectively represent multivariate relationships and capture high-order interactions, making them widely applicable for modeling complex high-order dependencies in brain networks. However, existing hypergraph-based methods fail to adequately account for dynamic inter-regional changes and pathological variations during disease progression. To address these limitations, this paper proposes a Lesion-guided Hyperedge Prioritization and Optimization network (LHPO) for early AD prediction, which enhances prediction accuracy by optimizing hypergraph structures based on inter-regional pathological differences. First, we design a lesion-aware brain region screening module that identifies AD-relevant regions through feature importance analysis. Second, a hyperedge optimization module incorporates lesion-based priors to adjust hyperedge weights during hypergraph construction, thereby refining the representation of inter-regional relationships. Finally, hypergraph convolution is employed for disease diagnosis, leveraging its high-order information aggregation capability to improve diagnostic accuracy and robustness. Experimental results demonstrate that LHPO achieves an accuracy (ACC) of 90.30% on baseline sMRI scans from the ADNI dataset. Furthermore, by analyzing functional compensation and pathological spread mechanisms across brain regions based on lesion-guided hypergraph structures, our framework provides theoretical insights for early intervention and treatment strategies.