Tracking the Transitions of Brain States: An Analytical Approach Using EEG

Abstract

Classification of the neural activity of the brain is a well known problem in the field of brain computer interface. Machine learning based approaches for classification of brain activities do not reveal the underlying dynamics of the human brain. Since eigen decomposition has been found useful in a variety of applications, we conjecture that change of brain states would manifest in terms of changes in the invariant spaces spanned by eigen vectors as well as amount of variance along them. Based on this, our first approach is to track the brain state transitions by analysing invariant space variations over time. Whereas, our second approach analyses sub-band characteristic response vector formed using eigen values along with the eigen vectors to capture the dynamics. We have taken two real time EEG datasets to demonstrate the efficacy of proposed approaches. It has been observed that in case of unimodal experiment, invariant spaces explicitly show the transitions of brain states. Whereas sub-band characteristic response vector approach gives better performance in the case of cross-modal conditions. Evolution of invariant spaces along with the eigen values may help in understanding and tracking the brain state transitions. The proposed approaches can track the activity transitions in real time. They do not require any training dataset.