Creating dMRI Biomarkers and Parsing Heterogeneity by Combining Multisite Datasets: Application to Autism Spectrum Disorder

Loading...
Thumbnail Image
Degree type
PhD
Graduate group
Bioengineering
Discipline
Engineering
Medical Sciences
Subject
autism spectrum disorder
big data
diffusion MRI
graph theory
machine learning
structural connectivity
Funder
Grant number
License
Copyright date
01/01/2025
Distributor
Related resources
Author
Shen, Rui, Sherry
Contributor
Abstract

The advent of precision medicine has revolutionized clinical research and practice globally by leveraging subject-specific biomarkers to avoid ineffective therapies and minimize adverse effects. In neuroimaging, there is a growing focus on identifying biomarkers with greater specificity and sensitivity. Diffusion magnetic resonance imaging (dMRI), in particular, offers unique insights into the brain's microstructural and network-level connectivity organization, making it a promising tool for biomarker discovery. However, the limited sample sizes of individual studies and the heterogeneous nature of many brain disorders present significant challenges, often leading to inconsistent or contradictory findings. While collaborative efforts have produced large-scale datasets, the field still lacks robust analytical methods to effectively address site-related variability and interpersonal heterogeneity. This thesis introduces a suite of innovative analytical tools aimed at overcoming the challenges of dMRI-based biomarker identification. These tools harness the power of big data by integrating multi-site datasets and employing advanced computational techniques such as machine learning, normative modeling, and network-based approaches. We begin by developing new harmonization methods that facilitate the integration of multi-site dMRI data, ensuring a comprehensive representation of brain disorders to improve the generalizability of dMRI markers while minimizing site-related biases. Biomarkers will be derived at multiple levels, including the regional microstructural level and the whole-brain network level. This multidimensional characterization aims to capture the complexity and heterogeneity of dMRI-based patterns in brain disorders. We apply these methods to autism datasets, exploring how the derived biomarkers can enhance understanding and facilitate subtyping of the disorder. Ultimately, this work advances biomarker research, particularly for heterogeneous conditions, and contributes to more robust discoveries through collaborative research in the era of team science and big data.

Advisor
Verma, Ragini
Huang, Hao
Date of degree
2025
Date Range for Data Collection (Start Date)
Date Range for Data Collection (End Date)
Digital Object Identifier
Series name and number
Volume number
Issue number
Publisher
Publisher DOI
Journal Issue
Comments
Recommended citation