Space-by-time decomposition for single-trial decoding of M/EEG activity.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Analyzing multichannel time-varying neuroimaging signals, such as M/EEG, requires sophisticated methods for single-trial analysis.
• Existing techniques may not fully capture the complex spatiotemporal dynamics inherent in neural activity.

Purpose of the Study:

• To introduce a novel space-by-time M/EEG decomposition methodology for enhanced single-trial analysis.
• To demonstrate the method's effectiveness in representing underlying neural structures and decoding task-related information.

Main Methods:

• Developed a space-by-time M/EEG decomposition technique based on Non-negative Matrix Factorization (NMF).
• Applied the method to an EEG dataset from a visual categorization task.
• Incorporated a decoding analysis to assess the functional role of extracted components.

Main Results:

• The NMF-based decomposition successfully extracted three temporal and two spatial components, providing a compact representation of M/EEG signals.
• Decoding analysis revealed reliable identification of stimulus presentation and task difficulty using component combinations.
• The proposed method demonstrated more robust decoding performance compared to a sliding-window linear discriminant algorithm across participants.

Conclusions:

• The space-by-time decomposition offers a meaningful low-dimensional representation of single-trial M/EEG data.
• This methodology effectively captures relevant neural information and facilitates robust decoding of experimental conditions.
• The findings support the utility of NMF-based decomposition for advancing M/EEG analysis and understanding neural processes.