Feature Interpretability in Motor Imagery Brain Computer Interfaces: A Meta-Analysis Across Connectivity, Spatial...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) translate neural activity into commands for communication, control, and neurorehabilitation.
• Noninvasive BCIs face challenges balancing classification accuracy with interpretability of underlying neural mechanisms.
• Motor imagery (MI)-based BCIs are crucial for restoring motor function but require transparent decoding methods.

Purpose of the Study:

• To conduct a meta-analysis of feature interpretability across common methods in motor imagery (MI)-based BCIs.
• To investigate how network topology and spatial organization, specifically brain network lateralization, contribute to MI decoding.
• To compare the neurophysiological plausibility of different feature extraction techniques.

Main Methods:

• Meta-analysis of feature interpretability for power spectral density, common spatial patterns (CSP), Riemannian geometry, and functional connectivity.
• Evaluation across multiple electroencephalography (EEG)-based BCI datasets.
• Analysis of brain network lateralization in sensorimotor and frontal regions.

Main Results:

• Common spatial patterns (CSP) and Riemannian geometry methods demonstrated superior classification performance.
• Network lateralization analysis revealed stronger neurophysiological plausibility.
• Robust lateralization patterns were observed in sensorimotor and frontal regions contralateral to imagined movements.

Conclusions:

• Connectivity-based features offer a complementary approach to enhance the interpretability of MI-based BCIs.
• Findings support the development of more transparent and clinically relevant BCI systems.
• Understanding neural mechanisms through lateralization improves biological relevance and clinical applicability of BCIs.