Probabilistic Cognitive State Modeling (PCSM): Decoding dynamic brain states to derive emergent cognitive processing properties from task fMRI

Abstract

Studying flexible, adaptive transitions between cognitive tasks and serial–parallel processing under changing task demands has been central to understanding human cognition. Advances in neuroimaging analysis have improved the ability to link cognition with brain function, motivating methods that characterize dynamic brain activity to quantify emergent cognitive properties during task-based fMRI.

Probabilistic Cognitive State Modeling (PCSM) combines Finite Impulse Response modeling of BOLD activity with a Gaussian Mixture Model–Hidden Markov Model to infer recurring multivariate patterns of task-evoked BOLD responses across spatially distributed regions over time (“brain states”). From the resulting posterior structure, PCSM deterministically derives interpretable processing metrics, including serial–parallel deviation, cognitive demand, and serial bottleneck.

Data-informed generative simulations evaluated PCSM across systematically varied noise levels and transition regimes. Results show that PCSM reliably recovers latent structure (∼98 % state-alignment accuracy under known generative conditions) and produces stable parameter estimates across simulation regimes. Threshold analyses identify reliable boundaries between parallel, mixed, and serial processing modes and recover expected relationships among demand, and bottleneck.

Together, these results demonstrate that PCSM provides a principled framework for characterizing dynamic task-evoked processing architectures and estimating individual-level cognitive dynamics from task-based fMRI, supporting future investigation of cognitive processing constraints across tasks and populations.