Predictability of Future Sequential Brain States from Resting-State fMRI

Establish whether specific future sequential brain states in human resting-state fMRI, represented as time-indexed vectors of regional blood oxygen level-dependent signals, can be predicted from previously observed time points.

Background

The authors frame brain-state prediction as an autoregressive task using transformer architectures and note a critical uncertainty: whether future resting-state brain activity can be forecast from past observations. Addressing this question has practical implications, including reducing scan time, aiding vulnerable patient populations, and enabling brain–computer interface applications.

Their paper explores transformer-based approaches on HCP data and reports promising short-horizon predictions, but the statement highlights the broader unresolved question of predictability in resting-state brain dynamics.