CONSTRUCTION OF SPATIOTEMPORAL NEONATAL CORTICAL SURFACE ATLASES USING A LARGE-SCALE DATASET

Mapping developmental regionalization and patterns of cortical surface area from 29 post-menstrual weeks to 2 years of age

Surface area of the human cerebral cortex expands extremely dynamically and regionally heterogeneously from the third trimester of pregnancy to 2 y of age, reflecting the spatial heterogeneity of the underlying microstructural and functional development of the cerebral cortex. However, little is known about the developmental patterns and regionalization of cortical surface area during this critical stage, due to the lack of high-quality imaging data and accurate computational tools for pediatric brain MRI data. To fill this critical knowledge gap, by leveraging 1,037 high-quality MRI scans with the age between 29 post-menstrual weeks and 24 mo from 735 pediatric subjects in two complementary datasets, i.e., the Baby Connectome Project (BCP) and the developing Human Connectome Project (dHCP), and state-of-the-art dedicated image-processing tools, we unprecedentedly parcellate the cerebral cortex into a set of distinct subdivisions purely according to the developmental patterns of the cortical surface. Our discovered developmentally distinct subdivisions correspond well to structurally and functionally meaningful regions and reveal spatially contiguous, hierarchical, and bilaterally symmetric patterns of early cortical surface expansion. We also show that high-order association subdivisions, where cortical folds emerge later during prenatal stages, undergo more dramatic cortical surface expansion during infancy, compared with the central regions, especially the sensorimotor and insula cortices, thus forming a distinct central-pole division in early cortical surface expansion. These results provide an important reference for exploring and understanding dynamic early brain development in health and disease.

Intrinsic Patch-Based Cortical Anatomical Parcellation Using Graph Convolutional Neural Network on Surface Manifold

Automatic parcellation of cortical surfaces into anatomically meaningful regions of interest (ROIs) is of great importance in brain analysis. Due to the complex shape of the convoluted cerebral cortex, conventional methods generally require three steps to obtain the parcellations. First, the original cortical surface is iteratively inflated and mapped onto a spherical surface with minimal metric distortion, for providing a simpler shape for analysis. Then, a registration or learning-based labeling method is adopted to parcellate ROIs on the mapped spherical surface. Finally, parcellation labels on the spherical surface are mapped back to the original cortical surface. Despite great success, spherical mapping of the original cortical surface is inherently sensitive to topological noise and cannot deal with the impaired brains that violate spherical topology. To address these issues, in this paper, we propose to directly parcellate the cerebral cortex on the original cortical surface manifold without requiring spherical mapping, by leveraging the strong learning ability of the graph convolutional neural networks. Also, we extend the convolution to the surface manifold using the kernel strategy, which enables us to over-come the notorious shape difference issue (e.g., different vertex number and connections) across different subjects. Our method aims to learn the highly nonlinear mapping between cortical attribute patterns (on local intrinsic surface patches) and parcellation labels. We have validated our method on a normal cortical surface dataset and a synthetic dataset with impaired brains, which shows that our method achieves comparable accuracy to the methods using spherical mapping, and works well on cortical surfaces violating the spherical topology.

CONSTRUCTION OF 4D NEONATAL CORTICAL SURFACE ATLASES USING WASSERSTEIN DISTANCE

Spatiotemporal (4D) neonatal cortical surface atlases with densely sampled ages are important tools for understanding the dynamic early brain development. Conventionally, after non-linear co-registration, surface atlases were constructed by simple Euclidean average of cortical attributes across different subjects, which leads to blurred folding patterns and therefore hampers the reliability and accuracy when registering new subjects onto the atlases. To better preserve the sharpness and clarity of cortical folding patterns on surface atlases, we propose to compute the Wasserstein barycenter, which represents a geometrically faithful population mean under the Wasserstein distance metric, for the construction of 4D neonatal surface atlases. The Wasserstein distance considers two distributions as heaps of sand, and quantifies their distance as the least cost to move all sand particles from one distribution to reshape it into the other. In our case, comparing to the direct vertex-wise Euclidean average, the Wasserstein distance takes into account the alignment of spatial distribution of cortical attributes, thus is robust to potential registration errors during atlas building. Using this method, we constructed 4D neonatal cortical surface atlases at each week, from 39 to 44 postmenstrual weeks, based on a large-scale dataset with 764 subjects. Our 4D atlases show sharper and more geometrically faithful cortical folding patterns than the atlases built by the state-of-the-art method, thus leading to boosted accuracy for spatial normalization and facilitating early brain development studies.

SPHERICAL U-NET FOR INFANT CORTICAL SURFACE PARCELLATION

In human brain MRI studies, it is of great importance to accurately parcellate cortical surfaces into anatomically and functionally meaningful regions. In this paper, we propose a novel end-to-end deep learning method by formulating surface parcellation as a semantic segmentation task on the sphere. To extend the convolutional neural networks (CNNs) to the spherical space, corresponding operations of surface convolution, pooling and upsampling are first developed to deal with data representation on spherical surface meshes, and then spherical CNNs are constructed accordingly. Specifically, the U-Net and SegNet architectures are transformed to the spherical representation for neonatal cortical surface parcellation. Experimental results on 90 neonates indicate the effectiveness and efficiency of our proposed spherical U-Net, in comparison with the spherical SegNet and the previous patch-wise classification method.

Computational neuroanatomy of baby brains: A review

The first postnatal years are an exceptionally dynamic and critical period of structural, functional and connectivity development of the human brain. The increasing availability of non-invasive infant brain MR images provides unprecedented opportunities for accurate and reliable charting of dynamic early brain developmental trajectories in understanding normative and aberrant growth. However, infant brain MR images typically exhibit reduced tissue contrast (especially around 6 months of age), large within-tissue intensity variations, and regionally-heterogeneous, dynamic changes, in comparison with adult brain MR images. Consequently, the existing computational tools developed typically for adult brains are not suitable for infant brain MR image processing. To address these challenges, many infant-tailored computational methods have been proposed for computational neuroanatomy of infant brains. In this review paper, we provide a comprehensive review of the state-of-the-art computational methods for infant brain MRI processing and analysis, which have advanced our understanding of early postnatal brain development. We also summarize publically available infant-dedicated resources, including MRI datasets, computational tools, grand challenges, and brain atlases. Finally, we discuss the limitations in current research and suggest potential future research directions.