Machine-learning-based functional connectivity analyses: challenges and opportunities
Open Access
- Author:
- Gur, Shlomit
- Graduate Program:
- Neuroscience
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- July 31, 2019
- Committee Members:
- Vasant Gajanan Honavar, Dissertation Advisor/Co-Advisor
Vasant Gajanan Honavar, Committee Chair/Co-Chair
John Yen, Committee Member
Michele Diaz, Committee Member
Michael Nelson Hallquist, Outside Member
Kevin Douglas Alloway, Program Head/Chair - Keywords:
- machine learning
functional connectivity
fMRI
network neuroscience
longitudinal
multi-site - Abstract:
- Over the past decade, substantial progress has been made in neuroscience in general and in neuroimaging in particular. This progress has been propelled by many factors, including technological advances, the development of theory, methodology, and tools, and making data sharing a practice in neuroimaging research. Consequently, neuroimaging data have increased in volume and complexity, which according to some researchers marks a new era in the field, a big-data era. Coincidentally, one of the more prominent directions to have been espoused by researchers in the field is that of functional connectivity analysis and interpretation, using machine learning techniques. This direction, or sub-field, represents the intersection between three independent fields: neuroscience, network science, and machine learning. To date, the application of machine learning techniques to functional connectivity data has been dominated primarily by neuroscientists with limited expertise in machine learning and network science, and computer scientists with limited understanding of the data’s domain. However, we postulate that cross-talk between the fields is imperative for the contribution and progress of the sub-field. E.g., methods should be developed or adjusted to meet domain-specific needs, data in the domain should be curated to meet requirements of methods of interest, and proper use of both methods and data should be ensured. Against this background, this dissertation examines the current state of functional connectivity analyses, identifies challenges and opportunities, and addresses them with the application and development of domain-aware machine learning techniques. More specifically, the challenges and opportunities in the present dissertation pertain to: (i) population-condition interactions in static functional connectivity, (ii) multi-site static functional connectivity repositories, and (iii) evolving functional connectivity from different participants.