Time-resolved estimation of strength of motor imagery representation by multivariate EEG decoding

Area of Science:

• Neuroscience
• Brain-Computer Interfaces
• Signal Processing

Background:

• Multivariate decoding of brain activity offers high temporal resolution for accessing encoded information.
• Extracting the strength of neural representations within single trials remains an underexplored area.

Purpose of the Study:

• To investigate the feasibility of extracting motor imagery (MI) strength across time within single trials using multivariate decoding.
• To analyze the temporal dynamics of MI representations in electroencephalogram (EEG) data.

Main Methods:

• Applied Support Vector Machine (SVM) for multivariate decoding of MI representations from EEG.
• Conducted time-resolved, single-trial analyses of decoding performance during motor imagery tasks.
• Utilized a hierarchical genetic algorithm for feature selection to identify key EEG components.

Main Results:

• Cross-temporal decoding revealed dynamic and stationary phases of MI-relevant features.
• Identified contralateral alpha and beta frequency features over sensorimotor and parieto-occipital cortices as stationary MI-related patterns.
• Demonstrated that SVM prediction scores correlate with univariate feature amplitudes within single trials, reflecting MI strength variations.

Conclusions:

• Developed a robust method for estimating MI strength continuously within single trials.
• The findings have significant implications for advancing single-trial analysis techniques.
• This approach lays the foundation for improved, real-time neurofeedback in motor rehabilitation, reflecting dynamic MI strength.