Abstract
Brain signals recorded from the primary motor cortex (M1) are known to serve a significant role in coding the information brain-machine interfaces (BMIs) need to perform real and imagined movements, and also to form several functional networks with motor association areas. However, whether functional networks between M1 and other brain regions, such as these motor association areas, are related to the performance of BMIs is unclear. To examine the relationship between functional connectivity and performance of BMIs, we analyzed the correlation coefficient between performance of neural decoding and functional connectivity over the whole brain using magnetoencephalography. Ten healthy participants were instructed to execute or imagine three simple right upper limb movements. To decode the movement type, we extracted 40 virtual channels in the left M1 via the beam forming approach, and used them as a decoding feature. In addition, seed-based functional connectivities of activities in the alpha band during real and imagined movements were calculated using imaginary coherence. Seed voxels were set as the same virtual channels in M1. After calculating the imaginary coherence in individuals, the correlation coefficient between decoding accuracy and strength of imaginary coherence was calculated over the whole brain. The significant correlations were distributed mainly to motor association areas for both real and imagined movements. These regions largely overlapped with brain regions that had significant connectivity to M1. Our results suggest that use of the strength of functional connectivity between M1 and motor association areas has the potential to improve the performance of BMIs to perform real and imagined movements.
Introduction
Brain-Machine Interfaces (BMIs) typically rely on signals from the primary motor cortex (M1) to decode a user's intent. However, M1 does not act in isolation; it is part of a complex network involving motor association areas like the supplementary motor area (SMA) and premotor cortex (PMC). While we know these regions work together, it hasn't been clear whether the strength of the functional connections between them actually impacts how well a BMI can decode movement. If a user’s internal "wiring" between M1 and its supporting regions is weak, does that make their neural signals harder to translate into machine commands? Understanding this relationship is crucial for identifying why some users struggle with BMIs (a phenomenon sometimes called "BMI illiteracy") and for developing ways to improve performance through neurofeedback.
Methods and Materials
The researchers recruited 10 healthy volunteers (5 male, 5 female; mean age 29.8) and used Magnetoencephalography (MEG) to record brain activity during three types of right-hand movements: grasping, pinching, and elbow flexion. Participants performed these both as real physical actions and as imagined movements.
Key technical approaches included:
- Beamforming: To isolate signals, they extracted 40 "virtual channels" specifically within the left M1.
- Imaginary Coherence (IC): To measure functional connectivity, the team used IC in the alpha band (8–13 Hz). Unlike standard coherence, IC ignores signals with zero time-lag, which helps filter out "volume conduction" (the same signal being picked up by multiple sensors) and focuses on true interactions between brain regions.
- Neural Decoding: A Support Vector Machine (SVM) was trained on the M1 virtual channel data to classify which of the three movements the participant was performing or imagining.
- Correlation Analysis: Finally, they mapped the correlation between an individual's decoding accuracy and their M1-to-whole-brain connectivity strength.
Results
The study demonstrated that the strength of alpha-band functional connectivity between M1 and motor association areas is a significant predictor of BMI decoding accuracy. Individuals with more robust neural "communication" between these regions performed better at both real and imagined movement tasks.
Decoding Accuracy: * Real Movement: Mean accuracy was 67.1% (chance level is 33.3%). * Imagined Movement: Mean accuracy was 48.7%, which is lower than real movement but still significantly above chance.
Connectivity Mapping: For real movements, significant correlations between connectivity and accuracy were localized to the left PMC and postcentral gyrus. For imagined movements, the correlations were more distributed, involving the left parietal lobe and the right IFG. Interestingly, for imagined movements, connectivity with the right inferior frontal gyrus (IFG) correlated with better performance. This area is associated with motor inhibition, suggesting that "good" mental imagers are better at actively suppressing actual muscle movement while imagining.
Discussion and Conclusion
This paper highlights the alpha band (mu rhythm) as a potential gating mechanism. While high-frequency gamma waves are often the focus of BMI decoding, alpha oscillations modulate cortical excitability; the study suggests that strong alpha-band networking creates a more "stable" environment for motor information to be represented in M1. Furthermore, the brain regions where connectivity was most important for BMI performance (PMC, parietal areas) were the same regions that showed the strongest baseline connectivity to M1, suggesting that a "healthy" motor network is a prerequisite for high-performance BMIs.
The results indicate that by measuring a user's alpha-band connectivity during a pre-operative MEG scan, researchers might be able to predict how effective a BMI will be for that specific patient. Future work needs to verify if these alpha-band biomarkers hold true in clinical populations, such as patients with ALS or spinal cord injuries, who may have reorganized brain networks. If connectivity correlates with performance, users might even be trained via neurofeedback to increase their M1-PMC connectivity, potentially "leveling up" their ability to use a BMI.
Summary updated