Identifying the latent active patterns underlying the dynamic organization of human brain using resting-state fMRI

The dynamic organization of human brain functional networks can be revealed through resting-state fMRI. It is still an open question about how to determine the essential factors underlying the dynamic activity of human brain. In this study, we proposed the assumption that the dynamic activity of brain was companied with various involvements of latent active patterns (LAPs). We further supposed that LAPs were sparsely involved with different brain states. We modeled the assumptions by adopting a dictionary-learning framework. An online dictionary learning method was used to calculate the LAPs and the sparse loading parameters. Based on the results obtained from the resting-state fMRI dataset, we found some commonly represented LAPs that were involved with the default mode network, salience network and frontoparietal attention network. The sparse represented LAPs at each time point were related with the time-varying activity. LAPs provided a new viewpoint to mine the factors related with the dynamic organization of brain activity.