Nonlinear ICA of fMRI reveals primitive temporal structures linked to rest, task, and behavioral traits

• Published in: NeuroImage (Orlando, Fla.) Vol. 218; p. 116989
• Main Authors: Morioka, Hiroshi, Calhoun, Vince, Hyvärinen, Aapo
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.09.2020; Elsevier Limited; Elsevier
• Subjects: Behavioral traits; Brain - physiology; Brain mapping; Brain Mapping - methods; Cognitive science; Datasets; Functional magnetic resonance imaging; Humans; Image Processing, Computer-Assisted - methods; Intelligence; Intelligence - physiology; Local space-contrastive learning (LSCL); Magnetic Resonance Imaging - methods; Memory, Short-Term - physiology; Neural networks; Neural Networks, Computer; Neural Pathways - physiology; Neuroimaging; Nonlinear spatial independent component analysis (sICA); Rest - physiology; Resting-state functional magnetic resonance imaging (fMRI); Short term memory; Task Performance and Analysis; Temporal primitives; Time series; Unsupervised deep learning; Unsupervised deep learning; Local space-contrastive learning (LSCL); Nonlinear spatial independent component analysis (sICA); Resting-state functional magnetic resonance imaging (fMRI); Behavioral traits; Temporal primitives
• ISSN: 1053-8119; 1095-9572
• DOI: 10.1016/j.neuroimage.2020.116989

Accumulating evidence from whole brain functional magnetic resonance imaging (fMRI) suggests that the human brain at rest is functionally organized in a spatially and temporally constrained manner. However, because of their complexity, the fundamental mechanisms underlying time-varying functional networks are still not well understood. Here, we develop a novel nonlinear feature extraction framework called local space-contrastive learning (LSCL), which extracts distinctive nonlinear temporal structure hidden in time series, by training a deep temporal convolutional neural network in an unsupervised, data-driven manner. We demonstrate that LSCL identifies certain distinctive local temporal structures, referred to as temporal primitives, which repeatedly appear at different time points and spatial locations, reflecting dynamic resting-state networks. We also show that these temporal primitives are also present in task-evoked spatiotemporal responses. We further show that the temporal primitives capture unique aspects of behavioral traits such as fluid intelligence and working memory. These results highlight the importance of capturing transient spatiotemporal dynamics within fMRI data and suggest that such temporal primitives may capture fundamental information underlying both spontaneous and task-induced fMRI dynamics. •A new nonlinear feature extraction framework: local space-contrastive learning.•LSCL extracts distinctive nonlinear temporal structures hidden in time series.•LSCL identifies primitive temporal patterns organizing resting-state fMRI.•The primitive patterns are also present in task-evoked spatiotemporal responses.•The primitive patterns capture unique aspects of behavioral traits.