Limb state accounts for differences between motor imagery and action in motor cortex.

The motor cortex is involved not only in movement execution but also in motor imagery, a process leveraged by decoding algorithms for brain-computer interface (BCI) applications in individuals with severe motor impairments. Previous work has shown that population activity during execution and imagery occupies partially overlapping regions of neural state space while also engaging distinct subspaces unique to each motor state, suggesting that decoders trained in one condition may not generalize to the other. Moreover, movement execution likely includes neural representations of both motor output and proprioceptive feedback, which themselves may occupy distinct or overlapping regions of neural state space. To explore these distinctions, we studied two individuals with incomplete spinal-cord injuries and partial residual proximal arm function performing a center-out reaching task in three conditions: motor imagery, active execution, and passive movement. We found that decoders trained on neural activity from motor imagery failed to generalize to either active or passive movements. In contrast, decoders trained on active or passive movement activity generalized reciprocally. Population analysis revealed distinct dynamics depending on limb state and proprioceptive feedback, which could explain this lack of generalization. These results suggest that motor imagery engages motor cortical representations distinct from those recruited during actual movements, either actively or passively generated, with important implications for the design of BCI decoders.