Computational methods for analysis of resting state functional connectivity and their application to study of aging / Harini Eavani.

Format:

Book

Manuscript

Thesis/Dissertation

Author/Creator:

Eavani, Harini, author.

Contributor:

Davatzikos, Christos, degree supervisor.

Bassett, Dani S., degree committee member.

Fischl, Bruce, degree committee member.

Satterthwaite, Theodore D., degree committee member.

Smith, Stephen, degree committee member.

Yushkevich, Paul, degree committee member.

University of Pennsylvania. Department of Bioengineering, degree granting institution.

Language:

English

Subjects (All):

Penn dissertations--Bioengineering.

Bioengineering--Penn dissertations.

Local Subjects:

Penn dissertations--Bioengineering.

Bioengineering--Penn dissertations.

Physical Description:

xvii, 123 leaves : illustrations (some color) ; 29 cm

Production:

[Philadelphia, Pennsylvania] : University of Pennsylvania, 2016.

Summary:

The functional organization of the brain and its variability over the life-span can be studied using resting state functional MRI (rsfMRI). It can be used to define a "macro-connectome" describing functional interactions in the brain at the scale of major brain regions, facilitating the description of large-scale functional systems and their change over the lifespan. The connectome typically consists of thousands of links between hundreds of brain regions, making subsequent group-level analyses difficult. Furthermore, existing methods for group-level analyses are not equipped to identify heterogeneity in patient or otherwise affected populations.

In this thesis, we incorporated recent advances in sparse representations for modeling spatial patterns of functional connectivity. We show that the resulting Sparse Connectivity Patterns (SCPs) are reproducible and capture major directions of variance in the data. Each SCP is associated with a scalar value that is proportional to the average connectivity within all the regions of that SCP. Thus, the SCP framework provides an interpretable basis for subsequent group-level analyses.

Traditional univariate approaches are limited in their ability to detect heterogeneity in diseased/aging populations in a two-group comparison framework. To address this issue, we developed a Mixture-Of-Experts (MOE) method that combines unsupervised modeling of mixtures of distributions with supervised learning of classifiers, allowing discovery of multiple disease/aging phenotypes and the affected individuals associated with each pattern.

We applied our methods to the Baltimore Longitudinal Study of Aging (BLSA), to find multiple advanced aging phenotypes. We built normative trajectories of functional and structural brain aging, which were used to identify individuals who seem resilient to aging, as well as individuals who show advanced signs of aging. Using MOE, we discovered five distinct patterns of advanced aging. Combined with neuro-cognitive data, we were able to further characterize one group as consisting of individuals with early-stage dementia. Another group had focal hippocampal atrophy, yet had higher levels of connectivity and somewhat higher cognitive performance, suggesting these individuals were recruiting their cognitive reserve to compensate for structural losses. These results demonstrate the utility of the developed methods, and pave the way for a broader understanding of the complexity of brain aging.

Notes:

Ph. D. University of Pennsylvania 2016.

Department: Bioengineering.

Supervisor: Christos Davatzikos.

Includes bibliographical references.

OCLC:

961021882