EEG classification of imagined syllable rhythm using Hilbert spectrum methods

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Decoding imagined speech is a key challenge in brain-computer interfaces (BCIs).
• Electroencephalography (EEG) offers a non-invasive method for capturing brain activity.
• Rhythmic patterns in imagined speech production are not well understood in EEG data.

Purpose of the Study:

• To investigate if the rhythm of imagined syllable production can be decoded from high-density EEG data.
• To develop and evaluate a novel signal processing method for analyzing non-stationary EEG signals.
• To assess the feasibility of an EEG-based communication system for imagined speech.

Main Methods:

• Collected high-density EEG data from seven subjects imagining syllables in different rhythms.
• Applied a modified second-order blind identification (SOBI) algorithm for artifact removal and dimensionality reduction.
• Extracted joint temporal and spectral features using Hilbert-Huang transformation (HHT) on SOBI components.
• Classified the three distinct rhythms using the extracted features.

Main Results:

• Classification performance for all subjects was significantly above chance, demonstrating successful rhythm decoding.
• The SOBI-HHT-based method outperformed traditional averaging-based methods in temporal, spectral, and time-frequency domains.
• Promising inter-trial transfer results indicate robustness of the decoding method.

Conclusions:

• The rhythmic structure of imagined syllable production is detectable in non-invasive EEG recordings.
• The SOBI-HHT method provides accurate time-spectral representations for non-stationary EEG data.
• This research is a significant step towards developing EEG-based communication systems for imagined speech.