A Covert Attention–Augmented Motor Imagery Hybrid Paradigm for Brain–Computer Interfaces

Motor imagery (MI) is widely used in brain–computer interfaces (BCIs) and has been applied in neurorehabilitation to support motor recovery in stroke patients. As an intrinsic paradigm, MI requires no external stimulation and remains natural and user-friendly. However, decoding accuracy is often limited under short input durations and in cross-subject settings. Although many algorithmic approaches have been explored, performance improvement is constrained by the variability and instability of MI-related activity, particularly in individuals with limited MI experience. Enhancing paradigm design and increasing the robustness of EEG features are therefore essential for reliable MI decoding. A covert attention-augmented motor imagery (CAA-MI) paradigm is introduced as a hybrid BCI (HBCI) approach designed to enhance the neural representations engaged during MI. The paradigm integrates covert spatial attention with motor imagery in a stimulus-free manner to evoke more discriminative cortical patterns. An experiment involving 17 participants was conducted to compare CAA-MI with traditional MI (T-MI). A transformer-based multi-branch EEG fusion network (TMEF-Net) was employed for classification. CAA-MI achieved higher accuracy than T-MI in both intra-subject and inter-subject evaluations, with clear advantages under short decoding windows. With a 3-second input, intra-subject and inter-subject accuracies reached 89% and 81%, respectively. CAA-MI elicited broader activation over occipital–parietal and sensorimotor regions, providing more reliable EEG representations. These findings indicate that CAA-MI provides a more robust and efficient paradigm for rehabilitation-oriented BCIs.