Statistical Methods for characterization and classification of brain functional networks: with application to Philadelphia Neurodevelopmental Cohort Study Open Access

Wang, Yikai (2015)

Permanent URL: https://etd.library.emory.edu/concern/etds/v692t6502?locale=en
Published

Abstract

Functional magnetic resonance imaging(fMRI)-based Brain Network analysis has stimulated an enormous amount of interest in recent years. Network-oriented research on brain plays the key role in understanding the complex brain architecture and functional organization. Given its importance, mapping of brain functional connectivity is a highly challenging task due to the enormous number of connections in the brain and the many confounding factors that lead to difficulty in estimating the true functional connection. In our thesis, I develop two statistical methods for better characterization and classification of brain functional networks. In the first part, I present a statistical strategy for estimating partial correlation matrix for characterizing brain network. Compared to the commonly used correlation, partial correlation can provide more accurate assessment of direct connection by controlling the confounding effects from other brain regions. The proposed approach can overcome the major technical difficulties that have prohibited reliable estimation of the partial correlations in neuroimaging data. We applied the proposed partial correlation approach on a sample of 505 subjects from Philadelphia Neurodevelopmental Cohort(PNC) study to investigate the sex-related difference in brain connectivity. I also estimated the network using the standard correlation. The results showed that the partial correlation based network discovered more relatively significant differential modules compared with the correlation based network and tended to find more within-module differences while the correlation network discovered more between-module differences. Moreover, partial correlation network achieved a higher sex classification accuracy than correlation network based on cross validation. In the second part, I propose to classify brain networks in PNC study using symmetric-positive-definite(SPD) kernel based PCA method. This method provides a compact representation of the high-dimensional brain network connectivity. I applied SPD-PCA method on PNC study to classify male and female subjects based on their brain functional connectivity patterns. Our method achieved a high accuracy rate(73.27%) in the classification using only 60 features. In comparison, the standard approach based on the precision matrix had an accuracy rate of 69.70% using as many as 34980 features. In summary, these findings highlight the advantages of more efficient and advanced statistical tools in characterization and classification of brain functional networks.

Table of Contents

Chapter 1 Introduction. 1

Chapter 2 Materials and Methods. 5

2.1 Philadelphia Neurodevelopmental Cohort (PNC) Study and Description

2.2 Functional Magnetic Resonance Imaging (fMRI) Data Preprocessing

2.3 Brain Network Construction

2.4 Functional Brain Network Comparison : Correlation vs. Partial Correlation

2.4.1 Graph Construction : Correlation and Partial Correlation

2.4.2 The Proposed Procedure for Estimating Partial Correlation Matrix for Neuroimaging

Data

2.4.2.1 A Constrained L1 Approach (CLIME) to Sparse Precision Matrix

2.4.2.2 Partial Correlation Matrix

2.4.3 Gender Classification based on Functional Brain Network: Correlation vs. Partial Correlation

2.4.3.1 Edgewise Comparison

2.4.3.2 Multivariate Pattern Analysis

2.5 A Compact Representation Method for Classification of Brain Network

2.5.1 Principle Component Analysis (PCA)

2.5.2 Symmetric Positive Definite (SPD) Kernel based PCA method

2.5.3 Gender Classification of Functional Brain Network using Compact Representation Methods

Chapter 3 Experiments and Results. 16

3.0 Choosing Tuning Parameters in CLIME

3.1 Results of Comparing Full-correlated Graph vs. Partial-correlated Graph

3.1.1 Results of Edgewise Comparison

3.1.2 Results of Multivariate Pattern Analysis

3.2 Results of Using Compact Representation of the Conditional Graph

Chapter 4 Discussion. 21

4.1 Conclusions and Recommendations

4.2 Strengths and Limitations

4.3 Future Direction

Reference. 24

Appendix. 26

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Subfield / Discipline
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files