Robust Brain-Machine Interface Design Using Optimal Feedback Control Modeling and Adaptive Point Process Filtering

Abstract

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.

Rapid control and feedback rates enhance neuroprosthetic control

“…The BMI was controlled by ensembles of multi-unit activity from the primary- and pre-motor cortices (15–33 units). At the beginning of each day, decoders were typically first trained using neural activity recorded during passive observation of cursor movements and were then adapted using closed-loop decoder adaptation (CLDA) techniques to achieve proficient control163132 (see the ‘Methods' section and Supplementary Note 3). …”

Review: Human Intracortical Recording and Neural Decoding for Brain–Computer Interfaces

“…Rather than modeling neural behavior as a time-evolving rate-function, spike times may be seen as a realization of a point processes with an inhomogeneous (i.e. time-varying) conditional intensity function [35], [114]–[117]. Approaching spike trains in this way allows the experimenter to model how the intensity function varies as a function of: (1) the neuron’s own intrinsic spike history, (2) the behavior of other recorded neurons, and/or (3) external variables of interest [117].…”

“…Approaching spike trains in this way allows the experimenter to model how the intensity function varies as a function of: (1) the neuron’s own intrinsic spike history, (2) the behavior of other recorded neurons, and/or (3) external variables of interest [117]. Recent decoding approaches have used point process methods in both NHP [35], [116] and human subjects [117]. …”

“…An alternative to requiring batch-based technician supervision is to perform closed-loop decoder adaptation. Here, the parameters for the decoder are updated as the user is controlling the end-effector in real-time [6], [35], [36], [124], [135]–[137]. Using this approach, NHPs can rapidly gain control of a decoder using a randomly initialized set of decoder parameter settings within minutes [135].…”

“…For instance, algorithmic decoder advancements that incorporate novel neurophysiological insights are more likely to capitalize on the useable signals available through single cell recordings and optimize control [58], [124]. Decoding algorithms may develop strategies to continuously update parameters to deal with system noise [15], [35], [124], [135], [158]. Algorithms may be developed to determine what parameters the brain will be able to adapt to use [38], [40], and what features necessitate recalibration of the decoding parameters[6].…”