Representation learning is a core component in data-driven modeling of various complex phenomena. Learning a contextually informative representation can especially benefit the analysis of fMRI data because of the complexities and dynamic dependencies present in such datasets. In this work, we propose a framework based on transformer models to learn an embedding of the fMRI data by taking the spatiotemporal contextual information in the data into account. This approach takes the multivariate BOLD time series of the regions of the brain as well as their functional connectivity network simultaneously as the input to create a set of meaningful features that can in turn be used in various downstream tasks such as classification, feature extraction, and statistical analysis. The proposed spatiotemporal framework uses the attention mechanism as well as the graph convolution neural network to jointly inject the contextual information regarding the dynamics in time series data and their connectivity into the representation. We demonstrate the benefits of this framework by applying it to two resting-state fMRI datasets, and provide further discussion on various aspects and advantages of it over a number of other commonly adopted architectures.

Temporal networks have become increasingly pervasive in many real-world applications, including the functional connectivity analysis of spatially separated regions of the brain. A major challenge in analysis of such networks is the identification of noise confounds, which introduce temporal ties that are nonessential, or links that are formed by chance due to local properties of the nodes. Several approaches have been suggested in the past for static networks or temporal networks with binary weights for extracting significant ties whose likelihood cannot be reduced to the local properties of the nodes. In this work, we propose a data-driven procedure to reveal the irreducible ties in dynamic functional connectivity of resting-state fMRI data with continuous weights. This framework includes a null model that estimates the latent characteristics of the distributions of temporal links through optimization, followed by a statistical test to filter the links whose formation can be reduced to the activities and local properties of their interacting nodes. We demonstrate the benefits of this approach by applying it to a resting-state fMRI dataset, and provide further discussion on various aspects and advantages of it. Author Summary In this work we propose an optimization-based null model to infer the significant ties, meaning the links that cannot be reduced to the local strengths and properties of the nodes, from the dynamic functional connectivity network. We asses multiple aspects of this approach and demonstrate that it is adaptable to most temporal segmentation methods. We demonstrate that this approach provides several advantages such as taking into account the global information of the network. We also compare the proposed model with several commonly applied null models empirically and theoretically.