Functional Connectivity-Guided Band Selection for Motor Imagery Brain-Computer Interfaces
New method wields functional connectivity to slash computation without sacrificing accuracy.
Motor imagery brain-computer interfaces (MI-BCIs) rely on decoding user-specific neural rhythms, but standard filter banks treat all bands equally, ignoring individual physiology. A new proof-of-concept study from Natália Araújo do Carmo and Aarthy Nagarajan takes a smarter approach: functional connectivity (FC) metrics guide which frequency bands to keep. By calculating phase-based connectivity (wPLI, PLV, PLI) over four sensorimotor channels and ranking bands by hemispheric coupling differences, their method selects the top K bands for CSP feature extraction.
The results are striking. On two datasets (BCI Competition IV-2a with 9 subjects and OpenBMI with 54), FC-guided band selection reduced the number of CSP fits by 22.2% to 77.8% while staying within a 2% accuracy baseline equivalence zone. PLV proved the most aggressive reducer, focusing on mu and low-beta rhythms; wPLI showed superior robustness across sessions by minimizing volume conduction effects. This principled, interpretable framework makes BCIs faster to calibrate and more practical for real-time use.
- Functional connectivity (wPLI, PLV, PLI) on only 4 sensorimotor channels guides band selection.
- Reduces required CSP fits by 22.2% to 77.8% while maintaining <2% accuracy drop from full 9-band FBCSP.
- PLV prioritizes mu and low-beta bands for maximum dimensionality reduction; wPLI offers inter-session stability.
Why It Matters
Faster, more personalized BCI calibration means real-time mind-controlled devices are closer to practical deployment.