Boosting specificity of MEG artifact removal by weighted support vector machine

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Magnetoencephalogram (MEG) data is susceptible to artifacts that can compromise analysis.
• Existing artifact removal methods may struggle with imbalanced data or preserving essential neural components.

Purpose of the Study:

• To develop and validate an automatic artifact removal method for MEG signals.
• To improve the accuracy and reliability of artifact detection and removal in MEG data.
• To ensure the preservation of diagnostically relevant neural components during artifact removal.

Main Methods:

• Utilized independent components analysis (ICA) for signal decomposition.
• Employed a weighted support vector machine (SVM) to handle imbalanced independent components (ICs).
• Implemented a re-weighting scheme within the SVM to prioritize the preservation of useful MEG ICs.

Main Results:

• The proposed method accurately distinguished artifacts from neural components in a manually marked MEG dataset.
• Achieved a high preservation rate of 99.72% ± 0.67 for MEG ICs.
• Demonstrated a classification accuracy of 97.91% ± 1.39 for artifact identification.
• Cross-validation (leave-one-subject-out) yielded an average accuracy of 97.41% ± 2.14, indicating robustness across subjects.

Conclusions:

• The developed automatic artifact removal method is effective for MEG data.
• The approach demonstrates high accuracy and superior preservation of neural signals.
• The method exhibits insensitivity to individual differences, making it broadly applicable.