Combining Generalized Eigenvalue Decomposing with Laplacian Filtering to Improve Cortical Decoding Performance

Artifact removal is a key step toward designing real-world and efficient brain computer interfaces. Here we describe an automatic blind source separation algorithm applicable to real-time signal processing. The algorithm combines the generalized eigenvalue decomposition technique with Laplacian filtering to separate desired and undesired subspaces, exclude artifact sources and recover artifact-free cortical signals. The algorithm outperforms commonly used artifact removal methods in brain computer interfaces as measured by cortical decoding performance.