Design and Decoding of a Novel Steering Motor Imagery Paradigm for Driving Brain-Computer Interfaces.

This study proposes a motor imagery brain-computer interface (MI-BCI) experimental paradigm for right-hand steering movements in driving scenarios. The proposed paradigm aligns more closely with actual steering behavior at the action level, aiming to address the mismatch between conventional motor imagery and actual driving tasks. To validate the feasibility of this paradigm, motor imagery EEG data were collected from 12 participants using an eight-electrode low-cost OpenBCI device and systematically analyzed. Time-frequency analysis revealed that, during right-hand steering imagery, electrode C3 exhibited event-related desynchronization (ERD) in both the mu and beta bands, whereas electrode C4 also showed ERD in the beta band but exhibited event-related synchronization predominantly in the mu band. Spatial pattern analysis further indicated that common spatial pattern (CSP) yielded discriminative weight distributions over the central region associated with steering direction, suggesting that the paradigm can elicit neural pattern differences corresponding to steering direction. On this basis, baseline methods, including CSP, FBCSP, EEGNet, EEG-TCNet, and ATCNet were employed to evaluate the separability of left- and right-turn steering tasks performed with the right hand under this paradigm. To improve decoding performance, the SMB-TCSAN model was developed, integrating Sinc convolution, multi-branch temporal convolution, and a multi-head self-attention mechanism, achieving an average classification accuracy of 66.80% on the steering-direction dataset. This study preliminarily validates the feasibility of low-cost, few-channel EEG devices in steering motor imagery decoding, providing a conceptual design reference for future research on driving-assistive BCI systems.