Computational Cortical Surface Analysis for Study of Early Brain Development Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
  • Meng, Yu
    • Affiliation: College of Arts and Sciences, Department of Computer Science
  • The study of morphological attributes of the cerebral cortex and their development is very important in understanding the dynamic and critical early brain development. Comparing with conventional studies in the image space, cortical surface-based analysis provides a better way to display, observe, and quantify the attributes of the cerebral cortex. The goal of this dissertation is to develop novel cortical surface-based methods for better studying the attributes of the cerebral cortex during early brain development. Specifically, this dissertation aims to develop methods for 1) estimating the development of morphological attributes of the cerebral cortex and 2) discovering the major cortical folding patterns. Estimation of the Development of Cortical Attributes. The early development of cortical attributes is highly correlated to the brain cognitive functionality and some neurodevelopmental disorders. Hence, accurately modeling the early development of cortical attributes is crucial for better understanding the mysterious normal and abnormal brain development. This task is very challenging, because infant cortical attributes change dramatically, complicatedly and regionally-heterogeneously during the first year of life. To address these problems, this dissertation proposes a Dynamically-Assembled Regression Forest (DARF). DARF first trains a single decision tree at each vertex on the cortical surface, and then groups nearby decision trees around each vertex as a vertex-specific forest to predict the cortical attribute. Since the vertex-specific forest can better capture regional details than the conventional regression forest trained for the whole brain, the prediction result is more precise. Moreover, because nearby forests share a large portion of decision trees, the prediction result is spatially smooth. On the other hand, missing cortical attribute maps in the longitudinal datasets often lead to insufficient data for unbiased analysis or training of accurate prediction models. To address this issue, a missing data estimation strategy based on DARF is further proposed. Experiments show that DARF outperforms the existing popular regression methods, and the proposed missing data estimation strategy based on DARF can effectively recover the missing cortical attribute maps. Discovery of Major Cortical Folding Patterns. The folding patterns of the cerebral cortex are highly variable across subjects. Exploring major cortical folding patterns in neonates is of great importance in neuroscience. Conventional geometric measurements of the cortex have limited capability in distinguishing major folding patterns. Although the recent sulcal pits-based analysis provides a better way for comparing sulcal patterns across individuals of adults or older children, whether and how sulcal pits are suitable for discovering major sulcal patterns in infants remain unknown. This dissertation adapts a sulcal pits extraction method from adults to infants, and validates the spatial consistency of sulcal pits in infants, so that they can be used as reliable landmarks for exploring major sulcal patterns. This dissertation further proposes a sulcal graph-based method for discovering major sulcal patterns, which is then applied to studying three primary cortical regions in 677 neonatal cortical surfaces. The experiments show that the proposed method is able to identify the previously unreported major sulcal patterns. Finally, this dissertation investigates and verifies that the sulcal pattern information could be utilized to help DARF for better estimating cortical attribute maps.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Li, Gang
  • Styner, Martin
  • Shen, Dinggang
  • Niethammer, Marc
  • Lin, Ming
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2017

This work has no parents.