Automated speech artefact removal from MEG data utilizing facial gestures and mutual information

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Speech production involves complex neural dynamics crucial for human communication.
• Non-invasive neuroimaging techniques like MEG have advanced speech research.
• Speech artifacts in MEG data, caused by facial muscle activity, obscure neural information.

Purpose of the Study:

• To develop an automated pipeline for removing speech artifacts from magnetoencephalography (MEG) data.
• To improve the accuracy and efficiency of analyzing neural processes related to speech production.
• To overcome limitations of manual artifact removal, such as time consumption and inconsistency.

Main Methods:

• Utilized independent component analysis (ICA) for artifact isolation.
• Employed electromyography (EMG) from facial muscles to identify speech-induced artifacts.
• Applied mutual information (MI) to measure similarity between EMG and MEG data for component selection.

Main Results:

• Successfully developed an automated pipeline for speech artifact removal from MEG data.
• The proposed method efficiently and automatically identified and removed speech artifacts.
• Demonstrated a feasible approach for transparent evaluation of removed and preserved MEG data.

Conclusions:

• The automated pipeline significantly enhances the analysis of speech-related neural dynamics.
• This method offers a standardized and reproducible approach to artifact removal in MEG studies.
• Facilitates a deeper understanding of the neural basis of speech production and language disorders.