Topics in Data Integration Methods for Neuroimaging and Generalized Additive Mixed Models for Ambulatory Blood Pressure Curves and Psychosocial Stressors Público

Murden, Raphiel (Summer 2021)

Permanent URL: https://etd.library.emory.edu/concern/etds/5712m7865?locale=pt-BR
Published

Abstract

Data integration methods, e.g., Joint and Individual Variation Explained (JIVE), simultaneously explore and analyze similarities between two or more sets of measures captured on the subjects. JIVE estimates shared and unique subspaces, which can be challenging to interpret. Chapter 1 expands upon insights into AJIVE as a canonical correlation analysis of principal component scores. This reformulation, which we call CJIVE, provides an ordering of joint components, uses a computationally efficient permutation test for the number of joint components, and can predict subject scores for out-of-sample observations. Extensive simulations show that AJIVE and CJIVE tend to select the joint rank correctly when true total signal ranks are provided. Using JIVE to integrate functional and structural connectivity from the Human Connectome Project, we find that joint scores from the first of two components are associated with fluid intelligence.

   

CJIVE only improves interpretation for two datasets. Furthermore, it remains unclear how to interpret JIVE decomposition for a single subject. Chapter 2 proposes Probablistic JIVE (ProJIVE), a model-based method for conducting JIVE analysis. ProJIVE provides a subject-level interpretation of the JIVE framework by modeling subject scores as random effects. Simulation studies show that ProJIVE estimates scores and loadings as well or better than existing methods. We applied ProJIVE to brain morphometry and cognitive/behavioral measures from the Alzheimer's Disease Neuroimaging Initiative (ADNI), which revealed associations between subject scores and Alzheimer's diagnoses. Variable loadings show that measurements of cortical and subcortical volume are strongly related to cognition measures.

Chapter 3 examines the relationship between household financial responsibility and ambulatory blood pressure (ABP) among black women in metro Atlanta. Previous studies of ABP use either a summary measure or inflexible parametric models. However, these approaches may result in the loss of substantial variability or unnecessarily constrain profile shape. Furthermore, ABP profiles are non-linear in time. We use generalized additive mixed models (GAMMs) to estimate ABP profiles for participants who are primarily responsible for earning household finances versus those who are not. GAMMs enable the assessment of periods during which the groups differ significantly, which may lead to interventions to help prevent adverse cardiovascular events.

Table of Contents

1 Canonical JIVE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Statistical Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2.1 JIVE Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4

1.2.2 Using CCA to Interpret JIVE: CJIVE . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Simulation study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11

1.3.1 Simulations comparing JIVE methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11

1.3.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4 Joint Analysis of Structural and Functional Connectivity in HCP Data . .16

1.4.1 Human Connectome Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16

1.4.2 Dimension Selection and Joint and Individual Variation Explained 17

1.4.3 Subject scores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.4.4 Variable Loadings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.4.5 Reproducibility and prediction of new subjects . . . . . . . . . . . . . . . . . . . 21

1.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 

2 Probabilistic JIVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.2.1 The original JIVE decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28

2.2.2 Probabilistic JIVE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28

2.2.3 Model identifiability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2.4 Expectation-Maximization Algorithm for ProJIVE . . . . . . . . . . . . . . . . . 32

2.3 Simulation study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.3.1 Simulations comparing JIVE methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.3.2 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35

2.4 Joint Analysis of Brain Morphometry and Cognition in ADNI Data . . . . . 38

2.4.1 TADPOLE Challenge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.4.2 Dimension Reduction, Preprocessing, and Summary . . . . . . . . . . . . . . 39

2.4.3 Joint Subspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3 Generalized Additive Models of Ambulatory Blood Pressure Profiles . . . . . . . 47

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.2 Ambulatory Blood Pressure in MUSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.3 Statistical Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .59

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .101

About this Dissertation

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
Degree
Submission
Language
  • English
Research Field
Palavra-chave
Committee Chair / Thesis Advisor
Committee Members
Última modificação

Primary PDF

Thumbnail Title Date Uploaded Actions
Topics in Data Integration Methods for Neuroimaging and Generalized Additive Mixed Models for Ambulatory Blood Pressure Curves and Psychosocial Stressors () 2021-08-12 11:56:37 -0400

Supplemental Files

Thumbnail Title Date Uploaded Actions