Optimal Versus Approximate Channel Selection Methods for EEG Decoding With Application to Topology-Constrained...

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Channel selection is critical for accurate neural decoding in electroencephalography (EEG).
• Current heuristic and approximation methods lack optimality guarantees, leaving the performance gap unclear.
• Efficiently selecting optimal electrode placements is a significant challenge.

Purpose of the Study:

• To quantify the optimality gap of state-of-the-art channel selection methods in least-squares based neural decoding.
• To introduce a mixed-integer quadratic programming (MIQP) framework for optimal channel selection.
• To evaluate the performance of different selection strategies, including topology-constrained placement.

Main Methods:

• Reformulated channel selection as a mixed-integer quadratic program (MIQP).
• Utilized efficient MIQP solvers to determine optimal channel combinations for up to 100 channels.
• Quantified performance losses of sub-optimal methods against the MIQP-derived optimal solution.
• Incorporated topology constraints for electrode placement in neuro-sensor networks.

Main Results:

• A greedy channel selection based on utility metric showed minimal optimality gap in auditory attention decoding.
• Other greedy and L1-norm penalized methods exhibited significant performance losses compared to the optimal selection.
• The MIQP formulation effectively integrated topology constraints for electrode placement.
• Topology-constrained electrode placement was feasible even for large-scale problems (>100 channels).

Conclusions:

• The MIQP approach provides an accurate benchmark for evaluating EEG channel selection methods.
• Greedy utility-based selection offers a near-optimal and efficient strategy.
• MIQP facilitates topology-aware electrode placement for advanced neural decoding applications.