A radiomics-based method for studying seed-based voxel-wise morphological connectivity

Individualized morphological brain networks are increasingly used to study the human connectome. However, most existing approaches remain atlas-dependent and region-level, obscuring within-region heterogeneity and limiting spatial specificity. Here, a radiomics-based morphological connectivity (MC) framework was proposed to map individualized coordination patterns of neural circuits at the voxel level. We extracted 93 radiomic features from gray matter volume maps and estimated MC by correlating each voxel's radiomic feature vector with that of a given seed. Using three independent datasets, we evaluated the feasibility, reproducibility, functional relevance, and clinical value of this approach within the triple-network model [default mode network (DMN), central executive network (CEN), and salience network (SN)]. Our proposed framework successfully recapitulated canonical network patterns across all three networks and demonstrated high between-subject consistency and good-to-excellent within-subject test-retest reliability. Moreover, seed-based MC exhibited significant coupling with seed-based functional connectivity (FC) at both group and individual levels, with stronger correspondence in the CEN. Finally, in a pediatric autism spectrum disorder cohort, seed-based MC revealed network-specific alterations in the CEN and DMN that were divergent from seed-based FC alterations. Taken together, the radiomics-based framework for seed-based voxel-wise MC offers a novel, atlas-free, and scalable approach for mapping brain networks in healthy and diseased conditions with high reproducibility and reliability.