EMG prediction from motor cortical recordings via a nonnegative point-process filter.

A constrained point-process filtering mechanism for prediction of electromyogram (EMG) signals from multichannel neural spike recordings is proposed here. Filters from the Kalman family are inherently suboptimal in dealing with non-Gaussian observations, or a state evolution that deviates from the Gaussianity assumption. To address these limitations, we modeled the non-Gaussian neural spike train observations by using a generalized linear model that encapsulates covariates of neural activity, including the neurons' own spiking history, concurrent ensemble activity, and extrinsic covariates (EMG signals). In order to predict the envelopes of EMGs, we reformulated the Kalman filter in an optimization framework and utilized a nonnegativity constraint. This structure characterizes the nonlinear correspondence between neural activity and EMG signals reasonably. The EMGs were recorded from 12 forearm and hand muscles of a behaving monkey during a grip-force task. In the case of limited training data, the constrained point-process filter improved the prediction accuracy when compared to a conventional Wiener cascade filter (a linear causal filter followed by a static nonlinearity) for different bin sizes and delays between input spikes and EMG output. For longer training datasets, results of the proposed filter and that of the Wiener cascade filter were comparable.

Stimulus-driven changes in sensorimotor behavior and neuronal functional connectivity application to brain-machine interfaces and neurorehabilitation.

Normal brain function requires constant adaptation as an organism interacts with the environment and learns to associate important sensory stimuli with appropriate motor actions. Neurological disorders may disrupt these learned associations, potentially requiring new functional pathways to be formed to replace the lost function. As a consequence, neural plasticity is a critical aspect of both normal brain function as well as the response to neurological injury. A brain-machine interface (BMI) represents a unique adaptive challenge to the nervous system. Efferent BMIs have been developed, which harness signals recorded from a tiny proportion of the motor cortex (M1) to effect control of an external device. There is also interest in the development of an afferent BMI that would supply information directly to the brain (e.g., the somatosensory cortex-S1) via electrical stimulation. If a bidirectional BMI that combined these interfaces were to be successful, new functional pathways would be necessary between the artificial inputs and outputs. Indeed, stimulation of S1 that is contingent upon the consequences of motor command signals recorded from M1 might form the basis for artificial Hebbian associations not unlike those driving learning in the normal brain. In this chapter, we review recent developments in both efferent and afferent BMIs, as well as experimental attempts to understand and mimic the Hebbian processes that give rise to plastic changes within the cortex. We have used a rat model to develop the computational and experimental tools necessary to describe changes in the way small networks of sensorimotor neurons interact and process information. We show that by repetitively pairing the recorded spikes of one neuron with electrical stimulation of another or by repetitively pairing electrical stimulation of two neurons, we can strengthen the inferred functional connection between the pair of neurons. We have also used the dual-stimulation protocol to enhance the ability of a trained rat to detect intracortical microstimulation behavioral cues. These results provide an important proof of concept, demonstrating the feasibility of Hebbian conditioning protocols to alter information flow in the brain. In addition to their possible application to BMI research, techniques like this may improve the efficacy of traditional rehabilitation for patients with neurologic injury.

Enhanced detection threshold for in vivo cortical stimulation produced by Hebbian conditioning.

Normal brain function requires constant adaptation, as an organism learns to associate important sensory stimuli with the appropriate motor actions. Neurological disorders may disrupt these learned associations and require the nervous system to reorganize itself. As a consequence, neural plasticity is a crucial component of normal brain function and a critical mechanism for recovery from injury. Associative, or Hebbian, pairing of pre- and post-synaptic activity has been shown to alter stimulus-evoked responses in vivo; however, to date, such protocols have not been shown to affect the animal's subsequent behavior. We paired stimulus trains separated by a brief time delay to two electrodes in rat sensorimotor cortex, which changed the statistical pattern of spikes during subsequent behavior. These changes were consistent with strengthened functional connections from the leading electrode to the lagging electrode. We then trained rats to respond to a microstimulation cue, and repeated the paradigm using the cue electrode as the leading electrode. This pairing lowered the rat's ICMS-detection threshold, with the same dependence on intra-electrode time lag that we found for the functional connectivity changes. The timecourse of the behavioral effects was very similar to that of the connectivity changes. We propose that the behavioral changes were a consequence of strengthened functional connections from the cue electrode to other regions of sensorimotor cortex. Such paradigms might be used to augment recovery from a stroke, or to promote adaptation in a bidirectional brain-machine interface.

Altering function in cortical networks by short-latency, paired stimulation.

Plasticity is a crucial component of normal brain function and a critical mechanism for recovery from injury. Numerous experimental studies have attempted to elucidate its underlying mechanisms under both in vitro and in vivo conditions. Short latency, associative pairing of presynaptic "trigger" spiking with stimulus-induced postsynaptic depolarization of a target neuron has been shown to lead to changes in the effectiveness of a stimulus applied to the presynaptic neuron. We have used similar methods to demonstrate changes in the statistically inferred functional connections among small groups of recorded neurons in rat sensorimotor cortex. These induced changes transcended simple changes in stimulus-evoked activity. Rather, they reflected a robust reorganization of network connectivity revealed by changes in the patterns of spikes in the cells' spontaneous discharge. We hypothesized that by strengthening the functional connections from trigger to target neurons, we might demonstrate a related behavioral change. To test this hypothesis, we trained rats to respond to a near-threshold, intracortical stimulus cue. Following 1-2 days of paired, short latency stimulation, the sensitivity of these rats to the cue was increased. The latency dependence and the timecourse of this effect were very similar to the corresponding parameters of the inferred connectivity changes in the first experiment. Such targeted connectivity changes may provide a tool for rerouting the flow of information through a cortical network, with profound implications for both rehabilitation and brain-machine interface applications.

Rewiring neural interactions by micro-stimulation.

Plasticity is a crucial component of normal brain function and a critical mechanism for recovery from injury. In vitro, associative pairing of presynaptic spiking and stimulus-induced postsynaptic depolarization causes changes in the synaptic efficacy of the presynaptic neuron, when activated by extrinsic stimulation. In vivo, such paradigms can alter the responses of whole groups of neurons to stimulation. Here, we used in vivo spike-triggered stimulation to drive plastic changes in rat forelimb sensorimotor cortex, which we monitored using a statistical measure of functional connectivity inferred from the spiking statistics of the neurons during normal, spontaneous behavior. These induced plastic changes in inferred functional connectivity depended on the latency between trigger spike and stimulation, and appear to reflect a robust reorganization of the network. Such targeted connectivity changes might provide a tool for rerouting the flow of information through a network, with implications for both rehabilitation and brain-machine interface applications.

Bayesian inference of functional connectivity and network structure from spikes.

Current multielectrode techniques enable the simultaneous recording of spikes from hundreds of neurons. To study neural plasticity and network structure it is desirable to infer the underlying functional connectivity between the recorded neurons. Functional connectivity is defined by a large number of parameters, which characterize how each neuron influences the other neurons. A Bayesian approach that combines information from the recorded spikes (likelihood) with prior beliefs about functional connectivity (prior) can improve inference of these parameters and reduce overfitting. Recent studies have used likelihood functions based on the statistics of point-processes and a prior that captures the sparseness of neural connections. Here we include a prior that captures the empirical finding that interactions tend to vary smoothly in time. We show that this method can successfully infer connectivity patterns in simulated data and apply the algorithm to spike data recorded from primary motor (M1) and premotor (PMd) cortices of a monkey. Finally, we present a new approach to studying structure in inferred connections based on a Bayesian clustering algorithm. Groups of neurons in M1 and PMd show common patterns of input and output that may correspond to functional assemblies.

Inferring functional connections between neurons.

A central question in neuroscience is how interactions between neurons give rise to behavior. In many electrophysiological experiments, the activity of a set of neurons is recorded while sensory stimuli or movement tasks are varied. Tools that aim to reveal underlying interactions between neurons from such data can be extremely useful. Traditionally, neuroscientists have studied these interactions using purely descriptive statistics (cross-correlograms or joint peri-stimulus time histograms). However, the interpretation of such data is often difficult, particularly as the number of recorded neurons grows. Recent research suggests that model-based, maximum likelihood methods can improve these analyses. In addition to estimating neural interactions, application of these techniques has improved decoding of external variables, created novel interpretations of existing electrophysiological data, and may provide new insight into how the brain represents information.

Biomimetic brain machine interfaces for the control of movement.

Quite recently, it has become possible to use signals recorded simultaneously from large numbers of cortical neurons for real-time control. Such brain machine interfaces (BMIs) have allowed animal subjects and human patients to control the position of a computer cursor or robotic limb under the guidance of visual feedback. Although impressive, such approaches essentially ignore the dynamics of the musculoskeletal system, and they lack potentially critical somatosensory feedback. In this mini-symposium, we will initiate a discussion of systems that more nearly mimic the control of natural limb movement. The work that we will describe is based on fundamental observations of sensorimotor physiology that have inspired novel BMI approaches. We will focus on what we consider to be three of the most important new directions for BMI development related to the control of movement. (1) We will present alternative methods for building decoders, including structured, nonlinear models, the explicit incorporation of limb state information, and novel approaches to the development of decoders for paralyzed subjects unable to generate an output signal. (2) We will describe the real-time prediction of dynamical signals, including joint torque, force, and EMG, and the real-time control of physical plants with dynamics like that of the real limb. (3) We will discuss critical factors that must be considered to incorporate somatosensory feedback to the BMI user, including its potential benefits, the differing representations of sensation and perception across cortical areas, and the changes in the cortical representation of tactile events after spinal injury.