EEG Microstate Sequences as Potential Brain-Computer Interface Triggers Derived from Motor Imagery Classification

EEG microstates, which are unique spatial distributions of brain activity, are thought to represent the building blocks of human thought. A recent study explores whether EEG microstate sequences could act as potential triggers for Brain-Computer Interfaces (BCIs). The researchers employed a semi-supervised deep learning model, specifically an LSTM-based autoencoder and a dense neural network, to classify left- and right-hand motor imagery EEG data. This classification output served as the BCI trigger.

The study was conducted using two approaches. Firstly, the autoencoder and classifier were trained separately. Secondly, an end-to-end approach was used, where training involved a combination of reconstruction and classification losses. The results indicated that the proposed model architecture could extract relevant features from microstate sequences and utilize them for within-subject and session classification.

Employing transfer learning for session-to-session or across-subject transfers led to peak classification accuracies around 89%.

The researchers also examined the extent to which transfer learning needed to be applied to achieve considerable classification accuracies, serving as the calibration time representative for the BCI. On average, around 400 seconds of calibration time were needed to reach 80% classification accuracy when using this approach. The study suggests that investigating EEG microstate trajectories could be a promising method for extracting BCI triggers.

By reducing the dimensionality of multi-channel recorded EEG signals to a distinct number of brain states over time, deep learning methods, particularly transfer learning, applied to EEG microstate trajectories appear promising for creating user-friendly and calibration-free BCIs in real-world applications.