Identifying and predicting EEG microstates with sequence-to-sequence deep learning models for online applications.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Electroencephalographic (EEG) microstates offer high-temporal-resolution insights into brain activity.
• Current clustering-based methods for EEG microstate analysis are offline and computationally intensive, limiting their application.
• Existing offline approaches are insufficient for cross-subject, cross-dataset, and multi-task scenarios.

Purpose of the Study:

• To develop a novel sequence-to-sequence framework for online EEG microstate identification and prediction.
• To enable end-to-end recognition and prediction from EEG signals to microstate labels.
• To provide a more efficient and adaptable approach for EEG microstate analysis.

Main Methods:

• Proposed a novel sequence-to-sequence framework for online microstate identification and prediction.
• Developed methods for constructing training datasets, including microstate label calibration, EEG electrode mapping, and sequence data partitioning.
• Validated the approach using four different neural network models on two public EEG datasets.

Main Results:

• Achieved cross-subject microstate recognition accuracy up to 74.26% for four microstates and 66.76% for seven microstates, outperforming KNN.
• Demonstrated prediction accuracy of 70.49% for four microstates and 62.71% for seven microstates.
• Confirmed that trainable models can effectively identify and predict EEG microstates.

Conclusions:

• Advanced EEG microstate analysis from an offline paradigm to an online model-data hybrid computation paradigm.
• The proposed framework offers new insights and references for cross-subject and cross-dataset EEG microstate applications.
• This approach enhances the feasibility and scope of utilizing EEG microstates in diverse research settings.