Computational methods for mapping the human cerebral cortex

Fijoy George Vadakkumpadan, Purdue University

Abstract

Human brain, the most complex living structure ever known, has been a subject of research for centuries. With the advent of revolutionary imaging techniques, neuroscientists can now acquire high-resolution structural and functional brain data in vivo. Integration of the acquired data using computational techniques to form maps of the human brain has extensive scientific and clinical applications. In this thesis, we develop novel computational methods for human brain mapping. Our focus is on the cerebral cortex, the most important component of the human brain. It is widely accepted that factors such as genetics, experience, and disease influence the shape of the cortical surface. These influencing factors can be indirectly studied by detecting and mapping the shape changes that they induce. The first major contribution of this thesis is a study which presents an effective yet simple method for detecting systematic cortical surface shape differences between different subject groups. This study has detected differences between Chinese and Caucasians, and the differences might be related to the genetic factor or the experiential factor, namely the language development. The second contribution is a new computational framework which unifies three computational procedures in cerebral cortex mapping, including cortical surface parameterization, cortical surface registration, and electromagnetic neural source imaging of the cortex. The parameterization represents the cortical surface in a form that facilitates the use of sophisticated mapping techniques. The registration integrates different cortical surfaces in a common coordinate space in which anatomically corresponding points occupy the same spatial coordinates. The source imaging identifies the electric currents in the cortex from electromagnetic measurements in vivo. Notably, the framework is based on the knowledge that the anatomy and physiology of the cortex is locally coherent. To make use of this local coherency, the framework adopts a representation scheme in which the cortical surface data are expressed using a linear combination of the low-frequency eigenvectors of a Laplacian of the underlying triangular mesh. Due to the conciseness and the faithfulness of this representation scheme, the computations are efficient as well as accurate.

Degree

Ph.D.

Advisors

Sun, Purdue University.

Subject Area

Neurosciences|Computer science

COinS