Deep Neural Network & Dynamic Functional Connectivity Analysis of Functional MRI Data

DEEP NEURAL NETWORK & DYNAMIC FUNCTIONAL CONNECTIVITY ANALYSIS OF FUNCTIONAL MRI DATA

Amaresh Kumar Mishra University of Houston-Clear Lake, 2022

Thesis Chair: Unal ‘Zak’ Sakoglu, PhD

This thesis work presents a dynamic functional connectivity (DFC)-based classification analysis of an already collected and completely de-identified functional magnetic resonance imaging (fMRI) dataset from two groups, veterans with Gulf War Illness (GWI), vs matched controls. Neuroimaging or brain imaging is the use of various techniques to either directly or indirectly image the structure, function, or pharmacology of the nervous system. fMRI is a neuroimaging technique which is used to measure brain activity by detecting changes associated with blood oxygenation level dependence (BOLD), which is an indirect measure of neural activity, and it helps obtain three spatial dimensional (3D) brain activation maps associated with certain stimulus and/or a task, depending on the experiments performed during the fMRI scan. Whole-brain resting-state fMRI (rsfMRI) data which were scanned from 23 GWI veterans (mean age 49.4) and 30 normal control (NC) veterans (mean age 49.8) were used for analyses. A computational method using DFC features, deep learning, and machine learning techniques were used to correctly classify GWI vs NC. Results show that, support vector machine (SVM) -based machine learning technique, combined with simple t-test method for feature extraction (using the DFC), performed better than convolutional neural network (CNN) deep learning method, in terms of classification accuracy (upwards of 98% accuracy for the former vs. upwards of 60% accuracy for the latter).