Movie viewing elicits rich and reliable brain state dynamics

The brain integrates dynamic sensory inputs with internal processes for adaptive functioning. Movie viewing, a naturalistic stimulus, requires dynamic processing of audiovisual content to form and update impressions and expectations. While dynamic analyses of functional neuroimaging data have advanced our understanding of brain activity patterns, capturing brain state dynamics and their role in perception and evaluation remains challenging. The brain exhibits coordinated activity changes even at rest, with dynamic patterns of functional brain connectivity potentially reflecting task-based phenotypes. Hidden Markov Models (HMMs) can model dynamic jumps between discrete brain states, revealing distinct states whose sequential expression yields observed fMRI data. Resting-state HMM analyses link discrete brain states to genetic and behavioral factors. However, resting-state variability and potential confounds limit inferences about behavioral relevance. Conventional task designs using abstract stimuli also impede the assessment of associations between stimulus processing and subjective evaluation (like engagement in a narrative). Naturalistic stimuli, such as movies, offer the replicability lacking in resting-state acquisitions and greater ecological validity than traditional task designs. Previous movie-viewing fMRI analyses using HMMs revealed a hierarchy of timescales, suggesting a remodeling of intrinsic correlations to match the statistical structure of naturalistic perceptual streams. Comparing resting-state and movie-viewing data using HMMs can characterize this remodeling process and investigate the functional significance of transitory brain states. This study uses HMMs to map brain states in fMRI data during resting state and movie viewing, assessing state validity using cross-session comparisons, movie annotations, the Neurosynth database, and physiological indices (heart rate and pupil diameter). It also investigates the correlation between brain state dynamics during movie watching and subjective immersion.

Previous research has explored dynamic patterns of functional brain connectivity, even in the absence of external tasks, using techniques like dynamic functional connectivity and hidden Markov models (HMMs). HMMs have proven useful in identifying discrete brain states during resting-state activity, linking these states to genetic and behavioral factors such as intelligence and personality. However, the variability inherent in resting-state neural dynamics and the influence of non-neural confounds like head motion limit the ability to draw firm conclusions about the behavioral relevance of these dynamic brain states. Studies using naturalistic stimuli like movies have shown promise in overcoming these limitations, offering greater ecological validity and allowing for the examination of the relationship between ongoing stimulus processing and subjective responses like engagement and interest. Previous work utilizing HMMs on movie viewing fMRI data has revealed a hierarchy of timescales in brain state transitions, mirroring the statistics of the natural world and suggesting that intrinsic brain activity is reshaped by naturalistic stimuli. The current study builds on this prior work by directly comparing resting-state brain dynamics to those observed during movie viewing.

Eighteen healthy participants underwent two fMRI scanning sessions, three months apart, during which they completed an 8-minute resting-state scan and a 20-minute movie-viewing session. Heart rate (HR) and pupil diameter (PD) were concurrently recorded. A post-movie questionnaire assessed subjective experience. The fMRI data was preprocessed using fMRIPrep and ICA-AROMA for motion correction and artifact removal. Data was filtered (0.01-0.15 Hz) and global WM and CSF signals were regressed. HR was preprocessed using FMRIB FastR and Tapas IO Toolbox; PD preprocessing was done using custom MATLAB scripts. A Hidden Markov Model (HMM) with ten states was used to analyze the fMRI data from 14 participants who completed both sessions. The HMM was fitted to concatenated time series from each participant, allowing for a direct comparison of states across conditions (rest and movie viewing, across sessions). The functional expression of the ten states was examined by coding their loadings onto 14 canonical brain networks. Neurosynth was used to reverse-infer the functional profiles of the brain states by associating their spatial expression with meta-analytic patterns of fMRI activity from the Neurosynth database. Inter-subject consistency of brain states during movie viewing was assessed using a reverse-inference approach, aligning state fluctuations with independent movie annotations (Table 1). Inter-session consistency was evaluated using the Jaccard index. Associations between brain states and physiological data (HR and PD) were analyzed, and the overlap between brain states and movie annotations (faces, scenes, language, changepoints) was calculated using the Szymkiewicz-Simpson overlap and permutation testing. Finally, inter-subject representational similarity analysis (IS-RSA) was used to examine the relationship between participant-specific brain state dynamics and their subjective movie ratings from the questionnaire.

The HMM analysis revealed ten distinct brain states with unique fMRI signal loadings across 14 canonical brain networks. Movie viewing was associated with significantly greater inter-subject consistency in brain state expression compared to rest (Figure 2). Specific movie events corresponded with high levels of inter-subject consistency in brain state occurrence (Table 1). Intersession consistency was significantly higher during movie viewing than during rest. Resting state was primarily characterized by bistable transitions between two states, while movie viewing elicited richer state transitions among a larger number of states. Neurosynth-based reverse inference revealed distinct functional profiles for each brain state (Figure 3), with states showing associations with various cognitive processes (e.g., language, emotion, sensorimotor functions, task-switching). Significant associations were found between brain states and both HR and PD (Table 2), supporting a link between movie-induced arousal and the emergence of brain states. Strong statistically significant associations were observed between specific brain states and movie annotations (Table 3), indicating temporal alignment between brain states and narrative features. The fractional occupancy (FO) of brain states differed significantly between rest and movie viewing (Figure 4). Network-based statistics revealed significant differences in state transition probabilities (Figure 5) between rest and movie viewing. Finally, IS-RSA revealed a correlation between participant-specific brain state dynamics (both FO and state transitions) and subjective movie ratings (Figure 7), suggesting a link between individual brain state patterns and subjective experience.

This study demonstrates that movie viewing, as a naturalistic stimulus, elicits richer and more reliable brain state dynamics than resting-state conditions. The observed brain states were associated with distinct functional profiles, physiological changes (HR and PD), and aspects of the movie narrative. The higher consistency of brain state dynamics during movie viewing compared to resting state highlights the potential of naturalistic paradigms for studying brain function. The significant correlation between individual brain state dynamics and subjective movie ratings underscores the functional relevance of these states in mediating subjective experience. These findings support the view that whole-brain state transitions reflect both exteroceptive (sensory input from movie) and interoceptive (internal bodily states) processes involved in the appraisal of ecologically valid sensory experiences. The use of a multi-session design and multimodal data enhances the robustness and generalizability of the findings.

This research demonstrates that movie viewing induces rich and reliable brain state dynamics that are temporally aligned with movie content, linked to physiological changes, and correlated with subjective appraisals. The study highlights the potential of naturalistic stimuli and HMM analyses for understanding brain dynamics in health and disease, particularly due to their higher test-retest reliability compared to resting state paradigms. Future research could explore the manipulation of specific movie features to evoke targeted brain states and further investigate the clinical implications of these findings.

The HMM is a statistical tool, not a biophysical model of neural activity, and it assumes discretely expressed brain states. The study used a relatively small sample size, although results were replicated across two sessions. While movie viewing mitigates some physiological confounds present in resting-state acquisitions, their effects cannot be completely eliminated. The relatively low temporal resolution of fMRI data may limit the analysis of fine-grained temporal relationships between brain states and movie events. The coarse-grained dimension reduction to 14 networks may have contributed to the restricted number of states observed at rest.