Invited Talk: Self-Supervised Learning for Longitudinal Neuroimaging Data
Speaker: Dr. Qingyu Zhao
Abstract
An increasing number of neuroimaging studies are acquiring longitudinal data to examine development in the structural and functional architecture of the human brain. This calls for the need for computational approaches that accurately model intra-subject dependencies within the repeated measures of the longitudinal design. In this talk, I will introduce several self-supervised deep learning approaches that leverage the repeated-measures design to derive meaningful and robust latent representations from raw 3D MRIs. Combining with the concept of factor disentanglement, these approaches can stratify changes and consistencies across multiple MRIs acquired from each individual over time. We applied these approaches to several longitudinal neuroimaging studies to highlight their strength in extracting the brain-age information from MRI, characterizing brain development in multi-modal imaging data, and revealing informative characteristics associated with neurodegenerative and neuropsychological disorders.
Biography
Dr. Zhao is a faculty in the Department of Psychiatry and Behavioral Sciences, Stanford University. He obtained his Ph.D. in computer science in 2017 from the University of North Carolina at Chapel Hill. His research has been focusing on identifying biomedical phenotypes associated with neuropsychiatric disorders by statistical and machine-learning-based computational analysis of neuroimaging and neuropsychological data. Dr. Zhao received a K99/R00 Pathway to Independence Award from National Institute on Alcohol Abuse and Alcoholism in 2021.