Consensus Matching Pursuit for multi-trial EEG signals.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Time-frequency representations are crucial for analyzing oscillatory brain activity in EEG, MEG, and intracranial EEG.
• Sparse time-frequency representations are gaining traction for describing complex neural data with fewer components.
• Existing multivariate Matching Pursuit (MP) algorithms often assume stable patterns across trials, which is not always true for brain signals.

Purpose of the Study:

• To adapt the Matching Pursuit algorithm for brain signals exhibiting cross-trial variability in time, frequency, and oscillation counts.
• To introduce a novel method, Consensus Matching Pursuit (CMP), that robustly handles inter-trial variability in neural data analysis.

Main Methods:

• Developed Consensus Matching Pursuit (CMP), an adaptation of the Matching Pursuit algorithm.
• Utilized a voting technique for robust atom selection, accommodating variability across trials.
• Adapted parameter fitting for each trial to account for individual trial characteristics.

Main Results:

• Validated CMP on simulated and real brain signal data, demonstrating robustness to variability.
• CMP provides more representative single-trial waveform estimates compared to existing multivariate MP algorithms.
• CMP achieves sparser data representations and enables quantification of cross-trial variability.

Conclusions:

• Consensus Matching Pursuit (CMP) is a robust and effective method for analyzing brain signals with significant cross-trial variability.
• CMP offers improved accuracy, sparsity, and variability quantification in time-frequency analysis of neural data.
• This method advances the analysis of complex neural dynamics, particularly in EEG and MEG studies.