Multimodal Graphical Network Analysis of Small-for-Gestational-Age in Preterm Infants: Integrating Neonatal Brain Volume, Structural Connectivity, and Early Neurodevelopmental Outcome

PURPOSE: Infants born Small-for-gestational-age (SGA) face heightened risks for cognitive and language impairments. The neurobiological mechanisms underlying these deficits remain unclear. This study aimed to identify multimodal neuroimaging biomarkers associated with fetal growth restriction and to characterize network-level associations linking SGA to early neurodevelopmental vulnerability using a data-driven graph-based framework. METHODS: In this prospective cohort of 186 preterm infants, near-term brain magnetic resonance imaging (MRI) and Bayley-III developmental assessments were analyzed. Multimodal imaging features-including volumetric indices from T2-weighted MRI and diffusion metrics from diffusion tensor imaging-were integrated with perinatal data. A sparse partial-correlation network was estimated using the Graphical Lasso algorithm (λ optimized via cross-validation) to infer conditional dependencies among features. Variables directly connected to birthweight Z-scores were identified as candidate biomarkers and validated for SGA classification and developmental outcomes using logistic regression and correlation analyses. RESULTS: Network analysis identified eight neuroanatomical correlates of birthweight Z-scores, including increased cerebrospinal fluid (CSF) volume; elevated axial, mean, and radial diffusivity in the left inferior longitudinal fasciculus (ILFL); higher axial diffusivity in the inferior fronto-occipital fasciculus; and altered degree centrality in the right precentral and posterior cingulate cortices (PCC). Logistic regression revealed CSF volume and ILFL diffusivity as independent predictors of SGA. Infants with language delay showed trend-level increases in ILFL diffusivity and reduced PCC centrality after FDR correction, suggesting possible associations between microstructural and connectomic alterations and language vulnerability. CONCLUSION: By integrating volumetric and diffusion MRI with graph-based modeling, this study uncovers latent neurobiological markers of SGA and provides clinically interpretable biomarkers for early risk stratification and individualized intervention.