Variable Selection of Neuroimaging Features in Mild Cognitive Impairment Public

Abstract

Mild cognitive impairment (MCI) is often a precursor to Alzheimer’s disease (AD), and early detection of MCI may facilitate treatment to prevent the onset of AD. One of the characteristics AD pathology is beta-amyloid (Aβ) plaques, which is detected using positron emission tomography (PET). However, PET imaging is expensive and invasive. We investigate whether low-cost, non-invasive structural magnetic resonance imaging (MRI) might be an alternative to detect brain atrophy and predict the subjects at risk of AD. Hundreds of features are typically derived from structural MRI, which makes variables selection difficult. We adopted a recently introduced method, Knockoffs filter, to select features from MRI while controlling the false discovery rate (FDR). To investigate the hypothesis of effectiveness of MRI in AD research, we have two main goals in our study: 1) to conduct three FDR-controlled methods of features selection between Aβ positive and negative for CN and MCI population; 2) evaluate feature selection procedures that predict MCI. The signals of feature for predicting MCI are much stronger than in predicting Aβ status. Although knockoff filter does not many features in Aβ pathology or in predicting MCI, some biologically plausible variables are selected in multiple initializations of the knockoff filter, which indicates left hippocampus volume is particularly important in predicting MCI. We propose to run many initializations of the knockoff-filter, which may improve the stability of feature selection.

Table of Contents

1 Introduction 1

2 Data and Methods 3

2.1 Features and Participants 3

2.2 Statistical Methods 5

3 Results 8

3.1 Aβ study 8

3.1.1 Univariate test results in Aβ study 9

3.1.2 Multivariate features results in Aβ study 10

3.1.3 Knockoffs features results in Aβ study 10

3.2 CN versus MCI study 13

3.2.1 Univariate test results in CN versus MCI study 13

3.2.2 Multivariate test results in CN versus MCI study 14

3.2.3 Knockoffs features results in CN versus MCI study 15

3.3 Coefficients estimation 16

4 Discussion 18

Appendix A Deleted MRI variables with missing percentage 21

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	Rollins School of Public Health
Department	Biostatistics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	M.S.P.H.
Submission	Master's Thesis
Language	English
Research Field	Statistics Health Sciences, Radiology
Mot-clé	High-dimensional feature selection Knockoffs filter Alzheimer’s disease False discovery rate
Committee Chair / Thesis Advisor	Risk, Benjamin , Emory University
Committee Members	Qiu, Deqiang, Emory University

Dernière modification

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Variable Selection of Neuroimaging Features in Mild Cognitive Impairment ()	2020-04-24 02:23:01 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions