Across-subjects classification of stimulus modality from human MEG high frequency activity

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Single-trial analyses reveal brain dynamics missed by averaging.
• Low signal-to-noise ratio (SNR) challenges single-trial methods.
• Decoding stimulus modality from brain activity is crucial for understanding sensory processing.

Purpose of the Study:

• To investigate the feasibility of single-trial magnetoencephalography (MEG) analysis for decoding stimulus modality (auditory vs. visual).
• To assess the effectiveness of combining beamformer source reconstruction with random forest classification for this task.
• To enable group-level inference through an across-subjects framework.

Main Methods:

• Applied beamformer source reconstruction to MEG data.
• Utilized random forest classification to distinguish auditory from visual stimuli.
• Integrated classification within an across-subjects framework for group inference.
• Analyzed high-frequency activity, particularly gamma band (75-95 Hz).

Main Results:

• Achieved good classification performance (66.44% accuracy across subjects) using single-trial gamma SNR.
• Demonstrated high consistency of high-frequency activity across trials and subjects.
• Identified informative activity in both auditory and visual cortex with high spatial specificity.
• Found gamma power most informative during stimulus presentation, with specific frequency bands (75-95 Hz, and 55-125 Hz in visual areas) being particularly discriminative.

Conclusions:

• Single-trial analysis of high-frequency MEG activity is feasible for decoding stimulus modality across subjects.
• Gamma band activity exhibits high trial-to-trial and inter-subject consistency, making it suitable for decoding.
• The study provides insights into the spatiotemporal and spectral characteristics of discriminative gamma activity relevant for sensory processing.