Abstract
BACKGROUND: Patients with congenital cardiac septal defect (CCSD) may still have neurodevelopmental abnormalities. This study aims to establish a combined model of brain magnetic resonance imaging (MRI) and clinical features for early prediction of neurodevelopment in infants with CCSDs. METHODS: Forty-eight infants diagnosed with CCSD by cardiac ultrasonography were prospectively enrolled. Neurodevelopmental outcome was assessed at 1 year of age using the Gesell Developmental Scale. Radiomics features of neonatal brain MRI and clinical characteristics at birth and follow-up were obtained. After the relationship between neurodevelopmental outcomes and clinical or radiomics features was assessed using Pearson and Spearman correlation analysis, we performed multiple stepwise linear regression models to determine the most significant predictors of developmental status. RESULTS: At 1 year of age, 25% (12/48) of infants with CCSD exhibited developmental delays, with the highest percentage of delays observed in language development (10.4%). Multiple stepwise linear regression analysis showed that shape features were independent predictors of gross motor (standardized coefficients β=-0.34; P=0.02). Guardians' education was an independent predictor of fine motor (standardized coefficients β=0.36; P=0.01). Wavelet-HLH-gray-level co-occurrence matrix-maximum correlation coefficient were independent predictors of adaptation (standardized coefficients β=-0.32; P=0.03). Wavelet-LHH-gray-level size zone matrix-small area low gray level emphasis (standardized coefficients β=0.31; P=0.03) and LoG-neighbourhood gray-tone difference matrix-strength (3D, σ=10 mm) (standardized coefficients β=0.29; P=0.03) were independent predictors of language development. Birth weight (standardized coefficients β=0.36; P<0.001) and wavelet-HLH-GLCM-cluster shade (standardized coefficients β=-0.27; P=0.04) were independent predictors of personal-social behavior. CONCLUSIONS: Some infants with CCSD may have neurodevelopmental delays at 1 year of age. A combined model based on radiomics features of neonatal brain MRI and clinical factors may be useful in predicting early neurodevelopment in infants with CCSD.