Ordinal Support Vector Classifier for Clinical Staging of Major Depression Using Multimodal Imaging Öffentlichkeit

Abstract

Ordinal Support Vector Classifier for Clinical Staging of Major Depression Using Multimodal Imaging

By Yujie Zhao

Introduction: Major depressive disorder (MDD) is a highly widespread, disabling, and pricey illness. Diagnosis and treatment of MDD is considered as a complex problem, because MDD results from a comprehensive interaction of social, psychological and biological factors. Patients received ineffective initial treatment would have significant personal and social costs as well as continued suffering. Identification of biomarkers of MDD in neuroimaging studies is a feasible method to improve diagnostic accuracy and will be helpful to guide treatment selection for individual patients. This study aimed to provide a neurobiological support for practical MDD diagnosis model - clinical staging model by establishing algorithms that discriminate clinical staging subtypes using machine-learning methodology and define most interesting features related to MDD clinical staging model to improve precision of diagnosis and treatment selection for MDD patients.

Methods: Three different treatment status (treatment naive, treatment responsive recurrent, treatment resistant) and control group were treated as a surrogate of clinical stage for MDD. Two ordinal multiclass support vector classifiers (SVM) were developed to classify subjects into these four clinical stages using functional magnetic resonance imaging data and diffusion tensor imaging data comparing to two traditional multiclass SVM classifiers. SVM recursive feature elimination (SVM-RFE) was applied after each model to select most significant features in this study.

Results: The result of cross-validation indicated that SVM models built on multimodal data have much better classification accuracy than those built on single neuroimaging modal. All-subset ordinal SVM model was more sensitive to ordinal features as well as similar classification accuracy compared to traditional one-to-one SVM model.

Discussion: With the hypothesis of ordinal trend in four clinical stage, all-subset ordinal model is more capable of defining most interesting features related to MDD clinical staging model. Therefore, this model could provide a new strategy using selected significant features for MDD early diagnosis and patients individualized treatment selection.

Table of Contents

Table of Contents

1. Introduction and Review of the Literature.....................................................................1

2. Methodology.............................................................................................................3

2.1. Study Overview......................................................................................................3

2.2. Objectives.............................................................................................................3

2.2.1. Objective 1: To Train Multiclass Classifiers for Four Ordinal Groups of Outcome.............3

2.2.2. Objective 2: To Define Most Significant Features Related to Four-Group Classification....4

2.3. Statistical Learning Methods.....................................................................................4

2.3.1. Support Vector Machine (SVM) ..............................................................................4

2.3.2. Multiclass SVM.....................................................................................................6

2.3.3. SVM Recursive Feature Elimination (SVM-RFE) ........................................................7

2.3.4. Cross-validation...................................................................................................8

2.4. Simulation Study ...................................................................................................8

2.5. Experimental Study.................................................................................................9

2.5.1. Dataset.............................................................................................................10

2.5.2. Screening..........................................................................................................11

3. Results...................................................................................................................11

3.1. Simulation Study...................................................................................................11

3.1.1 Classification Accuracy..........................................................................................11

3.1.2 Significant Features Selection................................................................................13

3.2. Experimental Study................................................................................................15

3.2.1 Screening...........................................................................................................15

3.2.2 Classification Accuracy..........................................................................................16

3.2.3 Significant Features Selection................................................................................18

4. Discussion...............................................................................................................19

References..................................................................................................................21

Appendices.................................................................................................................23

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 and Bioinformatics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	MPH
Submission	Master's Thesis
Language	English
Research Field	Health Sciences, Mental Health Health Sciences, Radiology
Stichwort	MDD Multiclass SVM Multimodal imaging
Committee Chair / Thesis Advisor	Guo, Ying, Emory University
Committee Members	Choi, Ki Sueng, Emory University
Partnering Agencies	Does not apply (no collaborating organization)

Zuletzt geändert

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Ordinal Support Vector Classifier for Clinical Staging of Major Depression Using Multimodal Imaging ()	2018-08-28 15:38:26 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions