Electrical and Computer Engineering ETDs

Publication Date

Fall 11-15-2016

Abstract

Mental disorders such as schizophrenia (SZ), bipolar (BD), and major depression disorders (MDD) can cause severe symptoms and life disruption. They share some symptoms, which can pose a major clinical challenge to their differentiation. Objective biomarkers based on neuroimaging may help to improve diagnostic accuracy and facilitate optimal treatment for patients. Over the last decades, non-invasive in-vivo neuroimaging techniques such as magnetic resonance imaging (MRI) have been increasingly applied to measure structure and function in human brains. With functional MRI (fMRI) or structural MRI (sMRI), studies have identified neurophysiological deficits in patients’ brain from different perspective. Functional connectivity (FC) analysis is an approach that measures functional integration in brains. By assessing the temporal coherence of the hemodynamic activity among brain regions, FC is considered capable of characterizing the large-scale integrity of neural activity.

In this work, we present two data analysis frameworks for biomarker detection on brain imaging with FC, 1) graph analysis of FC and 2) multimodal fusion analysis, to better understand the human brain. Graph analysis reveals the interaction among brain regions based on graph theory, while the multimodal fusion framework enables us to utilize the strength of different imaging modalities through joint analysis. Four applications related to FC using these frameworks were developed. First, FC was estimated using a model-based approach, and revealed altered the small-world network structure in SZ. Secondly, we applied graph analysis on functional network connectivity (FNC) to differentiate BD and MDD during resting-state. Thirdly, two functional measures, FNC and fractional amplitude of low frequency fluctuations (fALFF), were spatially overlaid to compare the FC and spatial alterations in SZ. And finally, we utilized a multimodal fusion analysis framework, multi-set canonical correlation analysis + joint independent component analysis (mCCA+jICA) to link functional and structural abnormalities in BD and MDD. We also evaluated the accuracy of predictive diagnosis through classifiers generated on the selected features. In summary, via the two frameworks, our work has made several contributions to advance FC analysis, which improves our understanding of underlying brain function and structure, and our findings may be ultimately useful for the development of biomarkers of mental disease.

Keywords

Functional connectivity, multimodal fusion, graph analysis, neuroimaging, biomarker, mental disorders

Document Type

Dissertation

Language

English

Degree Name

Computer Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Vince D. Calhoun

Second Committee Member

Vincent P. Clark

Third Committee Member

Marios S. Pattichis

Fourth Committee Member

Jing Sui

Download

Share

COinS