THIDS: Triplet High-Order Individual Differential Structural Brain Network

The human brain is a highly integrated network, and the brain network models serve as a simplified representation of this complex system. However, traditional brain network approaches are limited to group-level mean comparisons and first-order interactions, failing to capture the individual differences with higher-order interactions among brain regions in both brain development and heterogeneous psychiatric disorders. Here, we propose a triplet high-order individual differential structural brain network (THIDS), which is a linear combination of the differential (reference-perturbed) and perturbed networks, while utilizing the interaction information among all triplets of brain regions to construct an individual difference structured brain network. Results show that THIDS consistently achieves the highest accuracy in schizophrenia diagnosis, gender classification and cognition prediction across two datasets, four classification models and seven cognitive domains, compared with four alternative brain network construction methods. These results demonstrate that the proposed THIDS approach reliably captures individual differential features in the brain, showing great potential to investigate especially heterogeneous brain disorders and individual brain development.