Single-Trial Kernel-Based Functional Connectivity for Enhanced Feature Extraction in Motor-Related Tasks.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Motor learning involves brain plasticity and changes in neural network functional connectivity.
• Individual variability in motor skill acquisition is linked to differences in brain structure and function, often observed in electroencephalographic (EEG) recordings.

Purpose of the Study:

• To develop a kernel-based functional connectivity measure that addresses inter/intra-subject variability in motor-related tasks.
• To improve the evaluation of motor skills by analyzing brain activity patterns.

Main Methods:

• Extracted functional connectivity from spatio-temporal-frequency EEG patterns using Gaussian kernel cross-spectral distribution.
• Optimized spectral weights via a sparse ℓ2-norm feature selection framework for dimensionality reduction.
• Validated the method on three databases using motor imagery and motor execution tasks.

Main Results:

• The single-trial Gaussian functional connectivity measure achieved competitive classifier performance.
• The proposed method demonstrated reduced sensitivity to feature extraction parameters like sliding time windows.
• The approach avoided the need for prior linear spatial filtering and offered interpretability of connectivity patterns.

Conclusions:

• The kernel-based functional connectivity measure is robust and effective for analyzing motor-related brain activity.
• This novel metric shows significant potential for evaluating individual motor skills and understanding motor learning processes.
• The method offers a promising, less parameter-dependent alternative for brain connectivity analysis in motor tasks.