Using transient, effector-specific neural responses to gate decoding for brain-computer interfaces.

Real-world implementation of brain-computer interfaces (BCI) for continuous control of devices should ideally rely on fully asynchronous decoding approaches. That is, the decoding algorithm should continuously update its output by estimating the user's intended actions from real-time neural activity, without the need for any temporal alignment to an external cue. This kind of open-ended temporal flexibility is necessary to achieve naturalistic and intuitive control, but presents a challenge: how do we know when it is appropriate to decode anything at all? Activity in motor cortex is dynamic and modulates with many different types of actions (proximal arm control, hand control, speech, etc.), which can interfere with each other. Additionally, the "decodability" of any given action type (amount of relevant information present in the neural activity) fluctuates over time based on motor intent as well as intrinsic network dynamics. Here we present a method for simplifying the problem of continual decoding that uses transient, end effector-specific neural responses to identify periods of effector engagement. For example, we have observed unique neural signatures at the onset and offset of hand-related actions. Only after detecting the period of engagement do we then decode specific action features (e.g. digit movement or force). By using this gated approach, decoding models can be simpler (owing to local linearities) and are less sensitive to interference from cross-effector interference such as combined reaching and grasping actions. Clinical Trial ID: NCT01894802.