Support Vector Machine Classification of Resting State fMRI Datasets Using Clustered Dynamic Networks Pubblico

Byun, Hyo Yul (2014)

Permanent URL: https://etd.library.emory.edu/concern/etds/ng451h61q?locale=it
Published

Abstract

Resting state functional magnetic resonance imaging (rsfMRI) is a powerful tool for investigating intrinsic and spontaneous brain activity. Investigations on rsfMRI data are challenging due to its high dimensionality and the complex nature of brain functioning. The application of univariate and multivariate methods such as multi voxel pattern analysis have been instrumental in localizing neural correlates to various cognitive states and psychiatric diseases. However, many existing methods of rsfMRI analysis are insufficient for investigating the true mechanisms of the brain since they make implicit assumptions that are agnostic to the temporal and spatial dynamics of brain activity.

The proposed method in this thesis aims to create a superior feature space for representing brain activity and to create interpretable generalizations on these features for studying group differences by taking advantage of machine learning algorithms. k-means clustering is used to decompose dynamic resting state functional connectivity networks into discreet and holistic centroid networks. Next, the expression of these ideal centroid networks are computed for each subject and used as a new feature space for a support vector machine classifier. The interpretation of the generalizations computed by the SVM regarding the classification problem can be revealing since the feature space has been carefully designed to represent discreet dynamic brain states.

The novel method was tested for proof of concept using simulated and randomized data. Experiments show moderate success in classification ability for a real rsfMRI dataset including subjects with major depressive disorder and healthy controls. Although further research and optimization is necessary, the method holds promise as an investigative tool for studying group differences with rsfMRI and may hold future use in clinical applications.

Table of Contents

  • Introduction: 1
    • k-means Clustering: 2
    • Support Vector Machines: 4
    • Magnetic Resonance Imaging: 7
    • Limitations of MVPA and Univariate Statistical Analysis of fMRI: 11
    • Major Depressive Disorder: 14
  • Methods: 15
    • Subjects: 15
    • Image Acquisition: 16
    • rsfMRI Image Pre-Processing: 17
    • Regions of Interest Signal Extraction: 18
    • Collection of Dynamic Functional Connectivity Networks: 18
    • k-means Clustering of Dynamic FC Windows: 21
    • Support Vector Machine Training and Cross Validation: 23
    • Parameter Search: 25
  • Results: 26
    • Proof of Concept on Simulated Data: 26
    • Testing Resistance to Overfitting using Randomized Data: 30
    • Parameter Search on Subject Data: 32
    • Classification Results: 36
  • Discussion: 43
    • Two Way Disease State Classification (MDD vs. HC): 44
    • Three Way Disease State Classification (TRD vs. MDD vs. HC): 45
    • Recommendations for Future Studies and Limitations: 47
    • Conclusion: 51

About this Honors 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
Degree
Submission
Language
  • English
Research Field
Parola chiave
Committee Chair / Thesis Advisor
Committee Members
Ultima modifica

Primary PDF

Thumbnail Title Date Uploaded Actions
Support Vector Machine Classification of Resting State fMRI Datasets Using Clustered Dynamic Networks () 2018-08-28 10:22:22 -0400

Supplemental Files

Thumbnail Title Date Uploaded Actions