Stability of MEG for real-time neurofeedback

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Magnetoencephalography (MEG) offers real-time processing of movement-related field potentials for brain-machine interfacing (BMI).
• MEG's high sensitivity to magnetic fields results in a low signal-to-noise ratio, necessitating sufficient initial data for accurate motor activity characterization and decoder development.
• Efficient data collection is crucial for therapeutic BMI applications, enabling more time for neurofeedback training to promote cortical plasticity and rehabilitation.

Purpose of the Study:

• To determine the minimal amount of hand-grasp movement and rest data required for characterizing sensorimotor modulation depth.
• To evaluate the feasibility of building real-time brain state classifier functions using limited MEG data.
• To assess the accuracy of decoders trained on short, open-loop MEG data acquisition periods.

Main Methods:

• Collected five minutes of open-loop magnetoencephalography (MEG) data during hand-grasp movement and rest tasks.
• Utilized the collected MEG data to characterize sensorimotor modulation.
• Developed and trained classifier functions to decode brain states in real-time.

Main Results:

• Established that five minutes of initial open-loop MEG data is adequate for building effective decoders.
• Achieved real-time classification of brain activity as either grasp or rest with a high accuracy rate.
• Reported a classification accuracy of 84 ± 6% for distinguishing between grasp and rest states.

Conclusions:

• Limited initial data acquisition (five minutes) is sufficient for developing accurate real-time brain-machine interfaces using MEG.
• The findings support the potential for rapid deployment of therapeutic BMI systems, optimizing neurofeedback training.
• This study demonstrates the efficiency of MEG-based BMI for decoding motor intentions with minimal data.