No abstract

Resting state functional magnetic resonance imaging (rsfMRI) data exhibits complex but structured patterns. However, the underlying origins are unclear and entangled in rsfMRI data. Here we establish a variational auto-encoder, as a generative model trainable with unsupervised learning, to disentangle the unknown sources of rsfMRI activity. After being trained with large data from the Human Connectome Project, the model has learned to represent and generate patterns of cortical activity and connectivity using latent variables. The latent representation and its trajectory represent the spatiotemporal characteristics of rsfMRI activity. The latent variables reflect the principal gradients of the latent trajectory and drive activity changes in cortical networks. Representational geometry captured as covariance or correlation between latent variables, rather than cortical connectivity, can be used as a more reliable feature to accurately identify subjects from a large group, even if only a short period of data is available in each subject. Our results demonstrate that VAE is a valuable addition to existing tools, particularly suited for unsupervised representation learning of resting state fMRI activity.

No abstract

The architecture of the human connectome changes with brain maturation. Pivotal to understanding these changes is defining developmental periods when transitions in network topology occur. Here, using 110 resting-state functional connectivity data sets from healthy fetuses between 19 and 40 gestational weeks, we estimated optimal gestational-age (GA) cut points for dichotomizing fetuses into ‘young’ and ‘old’ groups based on global network features. We computed the small-world index, normalized clustering and path length, global and local efficiency, and modularity from connectivity matrices comprised 200 regions and their corresponding pairwise connectivity. We modeled the effect of GA at scan on each metric using separate repeated-measures generalized estimating equations. Our modeling strategy involved stratifying fetuses into ‘young’ and ‘old’ based on the scan occurring before or after a selected GA (i.e., 28 to 33). We then used the quasi-likelihood independence criterion statistic to compare model fit between ‘old’ and ‘young’ cohorts and determine optimal cut points for each graph metric. Trends for all metrics, except for global efficiency, decreased with increasing gestational age. Optimal cut points fell within 30–31 weeks for all metrics coinciding with developmental events that include a shift from endogenous neuronal activity to sensory-driven cortical patterns.