Identifying Event Boundaries in Naturalistic fMRI: Convergence across Activation, State, and Connectivity Methods

The brain dynamically responds to internal and external stimuli across multiple temporal scales. From this continuous stream of information, individuals automatically segment their experiences into meaningful, discrete events – a process fundamental to perception, memory, and learning. Event segmentation is hierarchically organized, such that fine-grained perceptual transitions are nested within broader, higher-order narrative or goal-directed units. This thesis asks how event segmentation is reflected in brain activity during film viewing, and whether distinct neuroimaging methods converge in identifying the same event boundaries. Using fMRI data from the FilmFestival dataset (OpenNeuro ds004042), this thesis compares methods for identifying event boundaries across two films of differing length, at three levels of analysis: activation, state, and connectivity. At the activation level, general linear models with hypothesis-driven, annotation-derived regressors test event boundaries across hierarchical levels. Inter-subject correlation (ISC) isolates the stimulus-driven signal shared across viewers, confirming the activation-level regions that are carried forward for analysis at the state level. At the state level, hidden Markov models and Greedy State Boundary Search detect when a region's multivoxel activity pattern reconfigures, segmenting each region into sequences of stable states. At the connectivity level, sliding-window correlation and wavelet coherence track how inter-regional coupling evolves over time and across frequencies. Finally, a convergence analysis tests whether the event boundaries identified from annotations, neural states, and connectivity align, and whether this agreement varies across the cortical hierarchy and across films. Grounded in Event Segmentation Theory, this comparison evaluates the suitability of activation-, state-, and connectivity-based methods for studying event-related brain dynamics, and positions film as a particularly revealing window onto how the mind segments continuous experience into discrete events.

Harkeerat Mangat is a MEiCogSci Student at the University of Vienna; Master’s thesis supervision by Priv.-Doz. Ronald Sladky.