Magnetoencephalographic artifact identification and automatic removal based on independent component analysis and...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Magnetoencephalography (MEG) signals are crucial for understanding brain activity but are often contaminated by artifacts from eye movements, heartbeats, and muscle activity.
• Artifact rejection leads to data loss and increased experimental time, potentially hindering analysis of noisy or highly contaminated MEG data.

Purpose of the Study:

• To develop and evaluate a robust method for isolating and removing artifactual components from MEG signals.
• To improve the efficiency and accuracy of artifact rejection in MEG data analysis.

Main Methods:

• Independent Component Analysis (ICA) was employed to decompose MEG signals into independent components (ICs).
• Threshold-based clustering, analyzing topographic, statistical, and spectral patterns of ICs, was used to identify artifactual components.
• An unsupervised neural network based on Adaptive Resonance Theory (ART-2) was also utilized for artifact IC categorization.
• Performance was assessed by measuring underestimation and overestimation of artifactual ICs.

Main Results:

• Threshold-based clustering successfully identified artifact-related ICs based on their distinct patterns.
• The combination of threshold-based clustering and ART-2 categorization yielded the best artifact identification performance.
• Comparison of pre- and post-processing MEG waveforms demonstrated the effectiveness of the proposed artifact rejection methods.
• Analysis indicated the potential for automatic artifact removal using general component templates.

Conclusions:

• The proposed combined approach of threshold-based clustering and ART-2 categorization offers a highly effective solution for artifact rejection in MEG.
• This method significantly improves data quality by minimizing data loss and preserving neural signal integrity.
• The findings suggest a pathway towards automated artifact removal in MEG analysis, enhancing research efficiency.