Spatiotemporal Filtering Bank for Decoding Motor Imagery EEG Signals

Common Spatial Patterns (CSP) have been extensively utilized in motor imagery brain-computer interfaces (BCIs). However, the frequency band and spatial specificity of subjects during motor imagery have limited the further development of CSP. To overcome these challenges, this paper introduces a novel algorithm called spatiotemporal filtering bank and classification (STFBC). Specifically, inspired by the concept of Filter Bank Common Spatial Patterns (FBCSP), we developed a spatiotemporal filter bank to extract fine-grained spatiotemporal features from EEG data. Subsequently, a heuristic feature selection method was employed to automatically identify a subset of features with low redundancy, high correlation, and complementarity for classification. To validate the effectiveness of the proposed algorithm, we conducted experiments using two public datasets: BCI Competition III dataset IVa and BCI Competition IV dataset IIa. In comparison to other competitive algorithms, the STFBC algorithm achieved the highest classification accuracy on these two datasets, with rates of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 6. 6 7 \%}$</tex> and 81.69 %, respectively. Furthermore, the spatial activation patterns revealed by STFBC were consistent with the brain regions activated during the motor imagery task.