Longitudinal Parcellation of the Infant Cortex Using Multi-modal Connectome Harmonics

Abstract

Functional segregation and specialization of cortical regions is central to the significant changes that take place during early brain development. We present an automated scheme that harnesses local and long-range connectivity features of the cortex—derived from multiple imaging modalities—for longitudinal parcellation of the early developing human cortex. This is realized by using multi-modal connectome harmonics in a hierarchical agglomerative clustering (HAC) framework to produce group correspondent, individual-specific, and age-dependent parcellation maps for the first two years of development. We observed decreased regularity in cluster size and shape with increasing time, reflecting cortical specialization known to emerge during early development. Investigation of the modularity of structural connectivity defined by our parcellations suggests convergence toward adult resting-state networks as the brain develops.