Functional fingerprinting for the developing brain using deep metric learning

Brain fingerprinting offers a promising method for delineating the unique functional architecture of individual brains using neuroimaging data. Here, we present a novel deep learning framework-Metric-BolT-for brain fingerprinting and use it to characterize distinct developmental trajectories during childhood and early adolescence. Based on longitudinal neuroimaging data, the extracted brain fingerprints achieved identification accuracies of 97.6% for two data runs acquired within a single session and 86.6% for runs spanning 4 years. Notably, the most discriminative brain fingerprints were driven by higher-order association cortices, particularly regions of the default-mode network. Moreover, we showed these fingerprints to correlate with cognitive abilities, such as fluid and crystallized intelligence, and to exhibit significant genetic associations, with stronger genomic relationships observed among individuals with more similar fingerprint patterns. Genes associated with brain fingerprinting tended to show upregulated expression in the frontal cortex, particularly in late childhood. Together, our study presents an innovative computational approach for brain functional fingerprinting and provides novel insights into individual variability in adolescent neurodevelopment.