Neuroplasticity may play a critical role in developing robust, naturally controlled neuroprostheses. This learning, however, is sensitive to system changes such as the neural activity used for control. The ultimate utility of neuroplasticity in real-world neuroprostheses is thus unclear. Adaptive decoding methods hold promise for improving neuroprosthetic performance in nonstationary systems. Here, we explore the use of decoder adaptation to shape neuroplasticity in two scenarios relevant for real-world neuroprostheses: nonstationary recordings of neural activity and changes in control context. Nonhuman primates learned to control a cursor to perform a reaching task using semistationary neural activity in two contexts: with and without simultaneous arm movements. Decoder adaptation was used to improve initial performance and compensate for changes in neural recordings. We show that beneficial neuroplasticity can occur alongside decoder adaptation, yielding performance improvements, skill retention, and resistance to interference from native motor networks. These results highlight the utility of neuroplasticity for real-world neuroprostheses.

The ability to decode mood state over time from neural activity could enable closed-loop systems to treat neuropsychiatric disorders. However, this decoding has not been demonstrated, partly owing to the difficulty of modeling distributed mood-relevant neural dynamics while dealing with the sparsity of mood state measurements. Here we develop a modeling framework to decode mood state variations from multi-site intracranial recordings in seven human subjects with epilepsy who self-reported their mood state intermittently over multiple days. We built dynamic neural encoding models of mood state and corresponding decoders for each individual and demonstrated that mood state variations over time can be decoded from neural activity. Across subjects, the decoders largely recruited neural signals from limbic regions, whose spectro-spatial features were tuned to mood variations. The dynamic models also provided an analytical tool to compute the timescales of the decoded mood state. These results provide an initial line of evidence indicating the feasibility of mood state decoding.

Neural activity exhibits dynamics that in addition to a behavior of interest also relate to other brain functions or internal states. Understanding how neural dynamics explain behavior requires dissociating behaviorally relevant and irrelevant dynamics, which is not achieved with current neural dynamic models as they are learned without considering behavior. We develop a novel preferential subspace identification (PSID) algorithm that models neural activity while dissociating and prioritizing its behaviorally relevant dynamics. Applying PSID to large-scale neural activity in two monkeys performing naturalistic 3D reach-and-grasps uncovered new features for neural dynamics. First, PSID revealed the behaviorally relevant dynamics to be markedly lower-dimensional than otherwise implied. Second, PSID discovered distinct rotational dynamics that were more predictive of behavior.Finally, PSID more accurately learned the behaviorally relevant dynamics for each joint and recording channel.PSID provides a general new tool to reveal behaviorally relevant neural dynamics that can otherwise go unnoticed.

Much progress has been made in brain-machine interfaces (BMI) using decoders such as Kalman filters and finding their parameters with closed-loop decoder adaptation (CLDA). However, current decoders do not model the spikes directly, and hence may limit the processing time-scale of BMI control and adaptation. Moreover, while specialized CLDA techniques for intention estimation and assisted training exist, a unified and systematic CLDA framework that generalizes across different setups is lacking. Here we develop a novel closed-loop BMI training architecture that allows for processing, control, and adaptation using spike events, enables robust control and extends to various tasks. Moreover, we develop a unified control-theoretic CLDA framework within which intention estimation, assisted training, and adaptation are performed. The architecture incorporates an infinite-horizon optimal feedback-control (OFC) model of the brain’s behavior in closed-loop BMI control, and a point process model of spikes. The OFC model infers the user’s motor intention during CLDA—a process termed intention estimation. OFC is also used to design an autonomous and dynamic assisted training technique. The point process model allows for neural processing, control and decoder adaptation with every spike event and at a faster time-scale than current decoders; it also enables dynamic spike-event-based parameter adaptation unlike current CLDA methods that use batch-based adaptation on much slower adaptation time-scales. We conducted closed-loop experiments in a non-human primate over tens of days to dissociate the effects of these novel CLDA components. The OFC intention estimation improved BMI performance compared with current intention estimation techniques. OFC assisted training allowed the subject to consistently achieve proficient control. Spike-event-based adaptation resulted in faster and more consistent performance convergence compared with batch-based methods, and was robust to parameter initialization. Finally, the architecture extended control to tasks beyond those used for CLDA training. These results have significant implications towards the development of clinically-viable neuroprosthetics.

No abstract

Motor function depends on neural dynamics spanning multiple spatiotemporal scales of population activity, from spiking of neurons to larger-scale local field potentials (LFP). How multiple scales of low-dimensional population dynamics are related in control of movements remains unknown. Multiscale neural dynamics are especially important to study in naturalistic reach-and-grasp movements, which are relatively under-explored. We learn novel multiscale dynamical models for spike-LFP network activity in monkeys performing naturalistic reach-and-grasps. We show low-dimensional dynamics of spiking and LFP activity exhibited several principal modes, each with a unique decay-frequency characteristic. One principal mode dominantly predicted movements. Despite distinct principal modes existing at the two scales, this predictive mode was multiscale and shared between scales, and was shared across sessions and monkeys, yet did not simply replicate behavioral modes. Further, this multiscale mode’s decay-frequency explained behavior. We propose that multiscale, low-dimensional motor cortical state dynamics reflect the neural control of naturalistic reach-and-grasp behaviors.

Brain-machine interfaces (BMI) create novel sensorimotor pathways for action. Much as the sensorimotor apparatus shapes natural motor control, the BMI pathway characteristics may also influence neuroprosthetic control. Here, we explore the influence of control and feedback rates, where control rate indicates how often motor commands are sent from the brain to the prosthetic, and feedback rate indicates how often visual feedback of the prosthetic is provided to the subject. We developed a new BMI that allows arbitrarily fast control and feedback rates, and used it to dissociate the effects of each rate in two monkeys. Increasing the control rate significantly improved control even when feedback rate was unchanged. Increasing the feedback rate further facilitated control. We also show that our high-rate BMI significantly outperformed state-of-the-art methods due to higher control and feedback rates, combined with a different point process mathematical encoding model. Our BMI paradigm can dissect the contribution of different elements in the sensorimotor pathway, providing a unique tool for studying neuroprosthetic control mechanisms.

Objective. Behavior is encoded across multiple spatiotemporal scales of brain activity. Modern technology can simultaneously record various scales, from spiking of individual neurons to large neural populations measured with field activity. This capability necessitates developing multiscale modeling and decoding algorithms for spike-field activity, which is challenging because of the fundamental differences in statistical characteristics and time-scales of these signals. Spikes are binary-valued with a millisecond time-scale while fields are continuous-valued with slower time-scales. Approach. We develop a multiscale encoding model, adaptive learning algorithm, and decoder that explicitly incorporate the different statistical profiles and time-scales of spikes and fields. The multiscale model consists of combined point process and Gaussian process likelihood functions. The multiscale filter (MSF) for decoding runs at the millisecond time-scale of spikes while adding information from fields at their slower time-scales. The adaptive algorithm learns all spike-field multiscale model parameters simultaneously, in real time, and at their different time-scales. Main results. We validated the multiscale framework within motor tasks using both closed-loop brain–machine interface (BMI) simulations and non-human primate (NHP) spike and local field potential (LFP) motor cortical activity during a naturalistic 3D reach task. Our closed-loop simulations show that the MSF can add information across scales and that the adaptive MSF can accurately learn all parameters in real time. We also decoded the seven joint angular trajectories of the NHP arm using spike-LFP activity. These data showed that the MSF outperformed single-scale decoding, this improvement was due to the addition of information across scales rather than the dominance of one scale and was largest in the low-information regime, and the improvement was similar regardless of the degree of overlap between spike and LFP channels. Significance. This multiscale framework provides a tool to study encoding across scales and may help enhance future neurotechnologies such as motor BMIs.