Prediction of early developmental status using neonatal structural MRI radiomics.
Authors
Affiliations (6)
Affiliations (6)
- Department of Scientific Research and Education, Jinjiang Municipal Hospital, Jinjiang, China.
- Department of Maternal and Child Health, School of Public Health, Peking University, Beijing, China.
- School of Public Health, Shandong Second Medical University, Weifang, Shandong, China.
- Fujian Key Laboratory of Neonatal Diseases, Xiamen Children's Hospital (Children's Hospital of Fudan University at Xiamen), Xiamen, China.
- Xiamen Key Laboratory of Neonatal Diseases, Xiamen Children's Hospital (Children's Hospital of Fudan University at Xiamen), Xiamen, China.
- Children's Hospital of Fudan University, National Children's Medical Center, Shanghai, China.
Abstract
Early identification of developmental vulnerability in high-risk neonates remains challenging because standardized developmental assessments, such as the Griffiths Development Scales-Chinese (GDS-C), are typically performed months after birth and cannot provide objective information during the neonatal period. Structural MRI provides quantitative information regarding neonatal brain development, and radiomics may capture imaging patterns beyond conventional visual assessment. This study aimed to investigate whether neonatal structural MRI radiomics combined with clinical variables could provide early estimates of developmental status during infancy. This prospective cohort study included 93 neonates recruited from Xiamen Children's Hospital between November 2022 and August 2025. All participants underwent non-sedated brain MRI on a 3.0-T scanner during the neonatal period, and radiomic features were extracted from structural MRI. MRI radiomic features were integrated with prespecified clinical variables to develop machine learning models for estimating the Griffiths Development Scales-Chinese (GDS-C) general quotient (GQ) obtained during follow-up. Model development and performance evaluation were performed using a nested cross-validation framework involving all participants. A two-level stacking framework integrating predictions from ten regression algorithms was developed, and predictive performance was evaluated using the coefficient of determination (R<sup>2</sup>), root mean square error (RMSE), and mean absolute error (MAE). Of the 93 neonates, 36 (38.7%) were female and 62 (66.7%) were preterm. The mean (SD) corrected age at GDS-C assessment was 7.13 (5.11) months, with a mean (SD) general quotient of 104.11 (19.77). The clinical-radiomics stacking model demonstrated moderate predictive performance under nested cross-validation, achieving an RMSE of 17.12, an MAE of 13.55, and an R<sup>2</sup> of 0.24. Compared with the clinical-only stacking model, incorporation of MRI radiomic features improved prediction performance. Exploratory analyses of GDS-C subdomains and hierarchical stacking approaches were additionally performed and should be interpreted as preliminary findings. Neonatal MRI radiomics combined with clinical variables may provide complementary information for estimating developmental status during infancy among high-risk neonates undergoing clinically indicated MRI. These findings support the potential of quantitative imaging approaches as adjunctive tools for early developmental surveillance rather than replacements for standardized assessments. Larger multicenter studies with longitudinal follow-up are required to further evaluate clinical utility.