Fig. 5: Temporal profiles of dynamic brain states are characteristic for individual subjects and region-network combinations.

To carve out the time scales of how distinct brain states click in and detach during movie watching, we quantified the dwell times of the states from each hidden Markov model (HMM) solution: we computed the average time of occupancy, between locking in and abandoning a given state, across the four states within an HMM solution at hand. We examined the 210 total trained HMM models in two ways: a Across subjects, the median value (the center bar in each box) ranges from 2.28 s to 25.31 s (x-axis was ordered from lowest to highest based on the median value, the same for (b)). The 15 subjects (assigned with random color) demonstrated distinct neural processing mechanisms tracking movie content. b Across region-network combinations (amygdala[AM]-network models in hash marked, hippocampus[HC]-network models in plane, the color is according to Schaefer-Yeo atlas³⁸), DN-tuned models with AM or HC coactivation partners showed the longest dwell time, while analogous VIS-tuned and LIM-tuned models showed the most consistently low dwell times with the smallest standard deviation. Again, the collective results witness the higher-order integration function of the DN to play a dominant role in movie watching. Boxplot: upper (lower) edge of the box is 25th (75th) percentile (interquartile distance); the middle line is the median value; the whiskers summarize the extreme data points of the distribution of median semantics-brain associations. Short names for Schaefer-Yeo networks: VIS Visual network, SMN somatomotor network, DAN dorsal attention network, DN default network, LIM limbic network, VAN salience and ventral attention network, FPC Frontoparietal network. Source data are provided as a Source Data file.