Multi-view learning based infant-specific cerebellum parcellation and its application to infants with early adversities

The National Institute of Child Health and Human Development awarded $427,625 to the University of North Carolina at Chapel Hill on July 15, 2026, under the Child Health and Human Development Extramural Research program (CFDA 93.865) to develop multi-view learning techniques for infant-specific cerebellum functional parcellation.

The project addresses a critical gap in neurodevelopmental research by creating advanced computational methods to map cerebellar functional organization during infancy. Existing parcellation approaches fail to account for the dynamic formation of cerebellar-cerebral connections in developing brains, rely on intra-cerebellar data alone without incorporating whole-brain interactions, and cannot capture developmental variability. The new methodology will enable researchers to characterize normal cerebellar maturation and detect functional deviations signaling early neurodevelopmental risk in infants exposed to early adversities.

Work is performed in Chapel Hill, North Carolina, with a period of performance through July 14, 2028. The award is classified as a Project Grant, the standard NIH research grant mechanism for investigator-initiated proposals. UNC-Chapel Hill, operating through its Office of Sponsored Research, serves as the prime recipient. The University of North Carolina System Office functions as the parent entity.

Abstract

The cerebellum is increasingly recognized as a critical hub not only for motor control, but also for cognitive and emotional development. Disruptions of cerebellar circuits in early life are strongly linked to widespread brain dysfunction and long-term neurodevelopmental disorders. Despite its importance, the cerebellum's functional organization during infancy and its role in early adversity-related developmental outcomes remains largely under- studied, leaving critical gaps in our understanding of normal developmental patterns of the cerebellum-involved whole brain connectome and how early adversities disrupt IT. A major barrier to advancing this research is the absence of infant-specific cerebellum functional parcellation, making IT nearly impossible to accurately characterize normal cerebellar maturation and detecting functional deviations that may signal early risk. Existing parcellation methods are unsuitable for infant cerebellum due to the following key limitations: 1) matching algorithms depend on mature cerebello-cerebral connections, which are still dynamically forming during infancy; 2) data-driven clustering focuses exclusively on intra-cerebellar connections, overlooking interactions with other brain regions; 3) conventional approaches typically target stable life stages, failing to account for developmental variability during infancy; and 4) while multi-view learning holds promise for modeling complex developmental dynamics across multiple circuits, existing methods treat all views equally and cannot capture the hierarchical structure across time and circuit dimensions. To address these methodology and knowledge gaps, this project aims to develop advanced multi-view learning techniques for infant cerebellum functional parcellation. This will not only enable a comprehensive understanding of cerebellar development but also facilitate the detection of early developmental abnormalities in the whole brain connectome perspective, particularly in infants with exposure to early adversities. Two specific aims are proposed. In aim 1, we will develop a cross-dimension shared-specific multi-view clustering (CSS-MVC) model that integrates multiple cerebellum-centered circuits across the brain while accounting for developmental dynamics over time, establishing the first set of both age-dependent and age-independent multi-scale infant cerebellum functional atlases. In aim 2, we will design an anatomy-aware graph convolutional network based multi-level anomaly detection (agcn-ad) method, which leverages large-scale infant neuroimaging data and the cerebellum-informed whole-brain connectome to assess the likelihood of abnormal brain development in individuals, particularly in infants exposed to early adversities, facilitating customized and targeted early interventions. Our computational models, code, and discoveries will be publicly released to the community to advance understanding of brain development, promote widespread adoption of cerebellum-informed analysis, and enable individualized early risk detection to improve neurodevelopmental outcomes in infants exposed to early adversities.