BCIwiki (bciwiki.com) — The study was published in the IEEE Journal of Biomedical and Health Informatics on August 10, 2026. Motor imagery-based brain-computer interfaces (MI-BCIs) have been extensively researched for neurorehabilitation and motor assistance, but online decoding performance of previous systems remains unsatisfactory.
Researchers Jing Y., Wang J., Que X., Li C., Liu S., and Xiang K. proposed a game theory-based adaptive human-machine joint (AHMJ) learning method integrating subject learning and decoder updating for MI-BCI decoding, successfully decoding multi-class MI for unilateral upper limb online. The MI training process was modelled as a two-player zero-sum minimax game, with task difficulty adaptively regulated according to each subject's performance. A new online adaptive algorithm integrating knowledge distillation and prototype-guided domain adaptation was designed for stable decoder updating.
In online MI-BCI experiments on fourteen healthy subjects and online simulation experiments on a 25-subject public dataset, average decoding accuracy was significantly improved by 9.7% and 5.6% respectively compared with traditional subject-only learning and previous human-machine joint learning methods (paired t-test, both p < 0.01). The online adaptive algorithm also outperformed previous updating approaches in both accuracy and stability.