BCI-sift: Automated feature selection toolbox boosts BCI decoding accuracy

Researchers from UMC Utrecht have released BCI-sift, a scikit-learn-compatible Python toolbox that automates feature selection for brain-computer interface (BCI) data. The toolbox tackles the challenge of high-dimensional, noisy neural signals by integrating advanced optimization algorithms to identify the most relevant features for machine learning tasks. It was validated on high-density electrocorticography (HD ECoG) data from eight able-bodied participants with 64-128 electrodes implanted over the sensorimotor cortex, who repeatedly spoke 12 words.

BCI-sift successfully identified informative neural features across electrode, temporal, and frequency dimensions. The selected electrode locations were consistent across participants and aligned with known functional organization of the sensorimotor cortex. Relevant time points clustered around speech production, and the high-frequency band was confirmed as most informative. Feature selection using BCI-sift improved classification accuracy compared to using all features. While validated on HD ECoG, the toolbox is broadly applicable to other BCI modalities, offering a versatile platform for improving decoding performance and interpretability.