OHBM 2023 | 2991 | Talk | Jeff Mentch | Identifying brain networks in a clinically rich and natura… скачать в хорошем качестве

OHBM 2023 | 2991 | Talk | Jeff Mentch | Identifying brain networks in a clinically rich and natura…

Title: Identifying brain networks in a clinically rich and naturalistic dataset using tensor decomposition. Session: Oral Session Speaker: Jeff Mentch Abstract: Much of our knowledge of cognition stems from experiments manipulating highly controlled stimuli. While this has proven fruitful, our everyday lives are richer and more dynamic. Recently, studies using naturalistic stimuli have gained increasing attention as a more ecologically valid method of studying the brain. While they have traditionally been used to examine the shared responses across groups of individuals, less work has targeted individual differences and heterogenous clinical populations like those with autism spectrum condition (ASC), characterized by altered audiovisual perception and social communication. Naturalistic stimuli are also amenable to data-driven approaches like independent component analysis (ICA), an influential tool for identifying brain networks that can be performed either individually or collapsing temporally or spatially for group analysis (Calhoun et al., 2009). This necessary concatenation of the data into a 2D representation may potentially lose some inherent low-rank structure shared among subjects. Further, the spatial or temporal independence assumed in the ICA model may not be biologically plausible given the extensive spatial and temporal overlap between brain networks. An alternative approach to this problem without imposing independence constraint is NASCAR, a stable and robust method for identifying brain networks and their temporal dynamics across subjects using tensor decomposition (Li et al., 2021). Here we apply NASCAR to fMRI data from naturalistic stimulation and at rest in a clinically rich dataset.