The human brain is rapidly growing during the last trimester and there is evidence that premature birth is related to adverse neurological events. Magnetic resonance (MR) imaging of premature neonates and subsequent automated segmentation allows us to quantify subtle disruptions in cortical growth and folding in large population based studies. We use an age-specific automated tissue segmentation technique combined with an age-specific parcellation scheme to study regional patterns of tissue growth in a large study of brain development in premature neonates. We aimed to investigate the additional information provided by regional analysis of MR scans when determining risks for delayed neurocognitive outcomes. Our image analysis framework incorporates a spatiotemporal reference atlas that covers the age range of 27.43 to 46.43 gestational weeks (GW), and consisted of manually segmented brain tissues and 8 hemispheric lobe regions. The tissue atlas was used to automatically segment all 269 MR scans, age range from 27.29 to 47.00 GW, using a hybrid framework that combines an age-specific atlas-based EM segmentation with a patch-based prior. This tissue segmentation was then used to re-align the subject anatomy with an age-specific tissue template and the region parcellation of each subject was carried out by template probability propagation into the subject anatomy from the spatiotemporal parcellation model. Multiple linear regression models of the parcellation volumes, divided into gray matter and white matter tissues, showed a significant correlation between regional volume and Bayley’s Scale of Infant and Toddler Development scores, beyond gestational age at birth. Parietal and occipital lobes were clear markers for language and motor abilities, while the temporal lobe volumes were correlated with cognitive abilities. These findings support a more anatomically-specific analysis of brain structure after premature birth and may provide the possibility of earlier and more specific clinical diagnostics and timely developmental interventions.

The brainstem's important role in the motor and sensory systems makes it a vital structure for studying pediatric development. However, the developing brain's rapid growth presents significant challenges when studying preterm infants. We investigated the regional differences of the brainstem between healthy and developmental delayed preterm infants. A spatiotemporal atlas was constructed from 28 training subject magnetic resonance (MR) scans and their brainstem manual tissue segmentations (whole brainstem) and parcellations (midbrain, pons, medulla) to capture a wide range of growth over time. The brainstem in all 269 MR scans in our database was automatically segmented using an age-specific template derived from the tissue-atlas training data to initiate an atlas-based Expectation-Maximization framework. We used tissue segmentations to re-register the age-specific tissue maps and derive an accurate non-rigid transformation from the atlas into the space of the subject scan. Template propagation was used to place an age-specific brainstem parcellation from the atlas into the space of the subject MRI. Multiple linear regression analysis was performed to model the relationship between regional brainstem volume calculated from the automated parcellations and specific Bayley Scales of Infant and Toddler development-Third Edition (BSID-III) cognitive, language, and motor composite scores at 18 months of age. We covaried with post-menstrual age at scan, age at birth, gender, and with and without supratentorial volume. Significant association existed for all brainstem volumes except for the pons and all BSID-III scores near birth age (mean=32.7+/-3.2 weeks) while there was nearly no significance for the same models in children near full-term age (mean=40.6+/-2.7 weeks). This supports the hypothesis that smaller total brainstem, midbrain, and medulla volumes measured shortly after preterm birth may be useful as a biomarker to predict poor developmental outcome at 18 months of age, while measurements later (near full term age) may be less useful as brain volume recovers.