Modeling Brain Development in Neonates

Post by Elisa Guma

What's the science?

During the perinatal period, the brain is rapidly developing, resulting in changes in size, gyrification, and contrast between tissues (as seen during brain imaging). To further complicate the situation, these changes may occur at different rates in different brain regions. This complexity makes it very challenging to accurately and reliably interpret clinical magnetic resonance imaging (MRI) data for those who may have experienced premature birth or perinatal brain injury. It is difficult to know what is abnormal for a neonatal brain given the age and clinical history of a patient. This week in Brain, O’Muircheartaigh and colleagues leveraged a large multi-contrast MRI dataset acquired during the perinatal period to model trajectories of normal brain development, and to accurately identify focal brain injury.  

How did they do it?

The authors used cross-sectionally acquired structural MRI data (T1- and T2- weighted) for 408 neonates, 189 of which were female, from the developing Human Connectome Project database, a publicly available dataset. The participants were all scanned postnatally, with post-menstrual ages ranging from 26 to 44 weeks (and gestational age at birth ranging from 23-42 weeks). A template was created from scans for 20 neonates over a wide age range based on two imaging features: T2-weighted image volume intensity and the cortical mantle. This was used to represent the midpoint for the sample’s age range. Next, the 408 scans were registered to the template, which means that the brain images were warped to match the template brain (using linear and nonlinear registration). The degree to which each voxel had to be expanded or shrunk to match the template gives us a measure of volume difference. The authors used a Gaussian process regression, which is a non-parametric approach, which they argue is a superior way to model growth curves for tissue intensity and shape at each voxel in the brain accounting for age (or degree of prematurity) and sex.

Next, the authors wanted to determine whether their model was longitudinally valid. Thus, for a subset of 46 neonates that had a second scan (excluded from the model construction), they quantified the deviation from the predicted image intensity. They then tested to see whether their model was able to detect deviations in tissue contrast that would predict the presence of punctate white matter lesions. A common brain injury associated with premature birth is a punctate white matter lesion, which is detectable using MRI. Their presence was identified in 40 neonates and manually labeled on each of the scans. Since focal abnormalities such as these lesions are reflected by deviations from typical development, the authors wanted to see if their model could accurately detect these deviations.

What did they find?

The authors were able to identify anatomically informed growth curves at each voxel in the brain, based on the MRI image intensity, with corresponding measures of variability, accounting for the degree of prematurity and sex of the infant. They observed differences in variability and shape in the growth curve based on structure; for example, the subventricular and intermediate zones were observed to have a rapid growth around term-equivalent age as they transition into white matter, whereas other white matter structures had a more linear growth, such as the sensory cortex. Interestingly, the frontal cortex had a flat curve in the perinatal period, as its development typically occurs later in life.

Longitudinal accuracy was also found to be high - 83% of the subset of neonates who had longitudinal scans had growth curves that matched those of the model. Accuracy, however, was not as good for younger neonates who had more intermediate structures present in their brains. The model also proved to have clinical utility as it was able to detect signal deviations due to punctate white matter lesions with high accuracy. 

What's the impact?

This study provides accurate estimates of non-linear changes in brain tissue intensity by modeling ex utero brain development over a wide age range (26 to 45 post menstrual weeks). Further, this work provides continuous growth charts for brain development based on shape and image intensity, similar to those used for height in clinical practice, providing an index that accounts for age and clinical history (i.e. prematurity). Future work may incorporate in utero MRI, and perhaps extend the postnatal period further.

Jonathan O’Muircheartaigh et al. Modelling brain development to detect white matter injury in term and preterm born neonates. Brain (2020). Access the original scientific publication here