← Projects · research

EEG microstate temporal dynamics

Tests whether including the duration of a brain state improves the model’s match to the recording. With nine states, including duration improves the match from 55.7% to 63.7%, compared with using only the current state. A separate comparison finds that its state-switching patterns differ more from chance than those from standard clustering (35.6% versus 23.6%).

This project uses deep learning to investigate EEG microstate temporal dynamics. It asks whether sequences contain structure beyond Markov and semi-Markov models, whether discrete labels discard meaningful continuous brain dynamics, whether dwell time shapes the next state, and whether switching rules remain stable across sessions.

These are research questions rather than reported outcomes. The case study introduces the question of whether brain states repeat.