Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.
Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.
STATISTICAL MACHINE LEARNING METHODS FOR MULTIVARIATE PATTERN ANALYSIS OF FMRI HIGH-DIMENSIONAL DATA
Author ORCID Identifier
https://orcid.org/0000-0001-5945-3758
AccessType
Campus-Only Access for Five (5) Years
Document Type
dissertation
Degree Name
Doctor of Philosophy (PhD)
Degree Program
Public Health
Year Degree Awarded
2022
Month Degree Awarded
September
First Advisor
Edward J. Stanek III
Second Advisor
Jing Qian
Third Advisor
Jacquie Kurland
Subject Categories
Biostatistics | Public Health
Abstract
Guidance is needed to choose the optimal algorithms to predict spoken word groups (Actions or Objects) from brain activation patterns using repeated fMRI (functional Magnetic Resonance Imaging) data. A set of predictive classifier models for categorical outcomes were evaluated including logistic regression, Lasso, Ridge Regression, Elastic Net, and Support Vector Machine (SVM), Gaussian Process (GP), Gradient Boosting (GB), Gaussian Naive Bayes (GNB), K-nearest Neighbors (KNN), and Random Forest (RF). The evaluation was conducted on simulated high-dimensional fMRI scenarios. Comparisons were made based on predictive performance (Accuracy, AUC, Precision, Recall, F1 Score, Time). Through the study design and simulation, we identified patterns between the predictive model performance and key fMRI parameters including Signal-to-noise Ratio (SNR), Spatial Resolution, Temporal Correlation, Number of Regions of Interest (ROIs), and Spacing between Regions. By comparison of different simulation scenarios, better prediction was possible with higher SNR, Spatial Resolution and Temporal Correlation. The best-performing algorithms were chosen by the model performance metrics and consistency, and included SVM and Elastic Net, followed by Logistic Regression and Ridge. RF and GB performed the worst. The best-performing classification algorithms from the simulation study were fitted to real-application fMRI data collected from healthy participants in a study of object and action picture naming in aphasia. The application evaluated the model performance and enabled the potential for clinically meaningful interpretation by identifying and visualizing multivariate voxel patterns.
DOI
https://doi.org/10.7275/30864108
Recommended Citation
Li, Minming, "STATISTICAL MACHINE LEARNING METHODS FOR MULTIVARIATE PATTERN ANALYSIS OF FMRI HIGH-DIMENSIONAL DATA" (2022). Doctoral Dissertations. 2651.
https://doi.org/10.7275/30864108
https://scholarworks.umass.edu/dissertations_2/2651
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.