Predicting Neurodevelopment in Very Preterm Infants Using Thalamic MRI Features and Machine Learning

Abstract

Preterm birth is the primary cause of infant death and is associated with later neurodevelopmental impairments. Neuroimaging is a powerful tool to analyse neuroanatomy abnormalities in preterm infants, allowing the examination of different brain structures, such as the thalamus and its alterations. Notably, the thalamus is a crucial hub for regulating cortical connectivity. Moreover, severe brain injury in preterm infants can impact thalamic growth and maturation over the long term. Therefore, studying thalamus morphology during the neonatal period using magnetic resonance imaging (MRI) can help identify markers that can predict neurodevelopmental outcomes in this vulnerable population. In this work, we semi-automatically segmented the thalamus structure from MRI scans and extracted thalamic morphological features from these segmentations. Consecutively, we employed K-means clustering, an unsupervised machine learning algorithm, to explore hidden patterns related to thalamic features in early and term-equivalent scans. Preliminary analyses revealed a significant association between these features and MRI scores used in clinical settings to assess brain development in very preterm infants at termequivalent age. The results showed 77% of preterm-born infants with abnormal MRI scores were clustered together, suggesting a strong link between these features and brain development complications. While several studies demonstrate the relationship between preterm birth and reduced thalamus volume at term-equivalent age, our study aims to investigate the link between thalamic volume trajectory during the early postnatal period and neurodevelopment at two years of age. Considering perinatal variables such as brain injury, we analysed the association of thalamic volume trajectory, calculated from early and term-equivalent MRIs of 116 very preterm infants, with cognitive, motor, and language outcomes at two years of age, assessed using the Bayley Scales of Infant and Toddler Development Third Edition. Our analysis employs bivariate methods to describe the study population in terms of clinical variables. In addition, we examined the impact of clinical variables on thalamic volume and then explored the relationship between thalamic volume and neurodevelopmental outcomes using multilevel linear regression models. Our results suggest an association between severe brain injury (p < 0.001) and reduced thalamic volume. Moreover, thalamic volume trajectory during early postnatal life was significantly associated with the three subscale scores of the neurodevelopmental assessment (cognitive: p = 0.004; motor: p = 0.001; language: p = 0.019). Finally, we employed Brain Age Gap Estimation (BrainAGE) based on thalamic volume to quantify delays in thalamic growth in very preterm infants compared to those born at term. Successively, we developed a machine learning classification model using BrainAGE analysis results, achieving 82% accuracy in predicting adverse neurodevelopmental outcomes. This model highlights the potential of the thalamus as a biomarker for the early identification of preterm infants at risk of adverse neurodevelopmental outcomes.