Dynamic pattern decoding of source-reconstructed MEG or EEG data: Perspective of multivariate pattern analysis and...

Area of Science:

• Neuroscience
• Brain-Computer Interface
• Biomedical Engineering

Background:

• Multivariate pattern analysis (MVPA) is increasingly used for decoding event-related responses from MEG/EEG sensor data.
• Applying MVPA to source-reconstructed MEG/EEG data is less common due to complexities like source orientation and signal leakage.
• Univariate analysis is the traditional method, but MVPA may offer advantages in capturing distributed patterns.

Purpose of the Study:

• To evaluate the effectiveness of MVPA for source-reconstructed MEG data, considering source orientation and signal leakage.
• To compare MVPA with univariate analysis for dynamic pattern decoding tasks.
• To propose and validate a signal leakage correction method for source-reconstructed data.

Main Methods:

• Used face vs. tool object category decoding (FvsT-OCD) on event-related responses from source-reconstructed MEG data.
• Compared univariate analysis and MVPA using single or multiple voxels.
• Implemented a novel symmetric signal leakage correction using independent component analysis.

Main Results:

• MVPA demonstrated higher sensitivity than univariate analysis for FvsT-OCD using single voxel information, leveraging all dipole orientations and reducing inter-subject variability.
• MVPA showed increased sensitivity when analyzing multiple voxels compared to a single voxel, indicating its ability to capture distributed patterns.
• Sensitivity significantly decreased after signal leakage correction, highlighting the impact of inter-regional signal leakage on MVPA outcomes.

Conclusions:

• MVPA is a sensitive approach for decoding brain activity from source-reconstructed MEG data, particularly when integrating information from multiple voxels.
• Source orientation information is effectively utilized by MVPA, enhancing its performance over univariate methods.
• Signal leakage between brain regions is a critical confound in MVPA of source-reconstructed data, necessitating correction for accurate interpretation.