Prediction of motor developmental outcomes based on MRI radiomics in premature infants.

Abstract

Background: This study aimed to combine radiomics and clinical variables to predict motor developmental outcomes in preterm infants.

Methods: Brain MRI images at term-equivalent age (TEA), perinatal clinical variables, and Bayley Scales of Infant and Toddler Development assessment results at 12-24 months corrected age were retrospectively collected for each preterm infant. Radiomic analysis was conducted on conventional MRI sequences, and a binary classification model for predicting motor neurodevelopmental outcomes was developed. A clinical prediction model was established on the basis of perinatal data, and a combined prediction model was created by integrating selected radiomic features. The predictive performance of each model was evaluated via the AUC.

Results: Preterm infants (n = 218) were included in the study, with 90 (41.3%) diagnosed with motor developmental delays. The AUCs of the prediction models were 0.75 (sensitivity 0.69, specificity 0.69) for multisequence radiomics, 0.87 (sensitivity 0.83, specificity 0.77) for clinical variables, and 0.94 (sensitivity 0.91, specificity 0.87) for the combined multisequence radiomics and clinical variables model.

Conclusion: A prediction model combining MRI-based radiomic biomarkers with clinical variables increases the accuracy of predicting motor developmental outcomes in preterm infants.

Impact: Our results showed that birth weight and maternal hypertensive disorders are important risk factors for motor nerve development abnormalities in preterm infants with no significant abnormalities on magnetic resonance imaging. The combination of multisequence radiomic Rad-scores and clinical variable models can improve the prediction efficiency of adverse prognoses. These findings contribute to the early detection of motor nerve abnormalities in related preterm infants and timely interventions for the benefit of families with preterm infants.