Electronic Thesis/Dissertation
 

Deep Generative Machine Learning Models for Brain Functional Connectivity Analysis of fMRI data

Open Access

This dissertation focused on the application of deep generative models for Functional Connectivity (FC) analysis of functional Magnetic Resonance Imaging (fMRI) data. The core idea of this application is to train deep generative models to synthesize neurotypical data. Therefore, when the model is applied to a neurodivergent data sample, it performs a translation of the neurodivergent sample to a neurotypical domain, assisting in the identification of atypical patterns. Using identified irregularities, we further perform FC analysis between large-scale brain networks. We test our framework on two distinct populations: individuals with Autism Spectrum Disorder (ASD) and Temporal Lobe epilepsy (TLE). This dissertation compares multiple generative ML models, including Variational Autoencoders and Denoising Diffusion Probabilistic Models. We perform an array of evaluation metrics that encompass the evaluation of the reconstruction of neurotypical data, evaluation of synthetic data generation, and assessment of fairness in identified impairments during FC analysis. Our results show that generative models are more effective in identifying spatial connectivity impairments in fMRI data rather than temporal ones. Moreover, identified impairments align well with those previously reported in the literature. The issue of bias in FC analysis, stemming from a significant imbalance in the data, has been highlighted in the previous literature. Therefore, in order to provide more targeted data synthesis and increase fairness in FC analysis, we advance our models by introducing a conditional mechanism, which allows for the addition of phenotypic information to the model. We demonstrate that a more targeted approach to data generation leads to improved reconstruction and synthesis, along with a reduction in bias during FC analysis.

Author Language Date created Type of Work License
  • All rights reserved
Rights statement GW Unit Degree Advisor Committee Member(s) Persistent URL

Notice to Authors

If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.

Thumbnail Title Date Uploaded Visibility Actions
Preview of Sidulova_gwu_0075A_16659.pdf Sidulova_gwu_0075A_16659.pdf 2024-01-11 Open Access