Multi-muscle FES force control of the human arm for arbitrary goals

Prediction of isometric forces from combined epidural spinal cord and neuromuscular electrical stimulation in the rat lower limb

Both epidural spinal cord and muscle stimulation have been widely used for restoration of movement after spinal cord injury. However, using both approaches simultaneously could provide more flexible control compared to using either approach alone. We evaluate whether responses evoked by combined spinal and muscle stimulation can be predicted by the linear summation of responses produced by each individually. Should this be true, it would simplify the prediction of co-stimulation responses and the development of control schemes for spinal cord injury rehabilitation. In anesthetized rats, we measured hindlimb isometric forces in response to spinal and muscle stimulation across a range of amplitudes. Force prediction errors were calculated as the difference between predicted co-stimulation vectors and observed co-stimulation vectors whereby small errors signified evidence for linear summation. We found that the errors for spinal and muscle co-stimulation were significantly larger than expected. Using a bootstrapping analysis, we find that these larger errors do not reflect a nonlinear interaction between spinal and muscle responses. Instead, they can be attributed to the variability of spinal stimulation responses. We discuss the implications of these results to the use of combined muscle and spinal stimulation for the restoration of movement following spinal cord injury.

Robot-Inspired Human Impedance Control Through Functional Electrical Stimulation

Functional Electrical Stimulation is an effective tool to foster rehabilitation of neurological patients suffering from impaired motor functions. It can also serve as an assistive device to compensate for compromised motor functions in the chronic phase occurring after a disease or trauma. In all cases, the dominant paradigm in FES applications is that of aiding specialized, task-specific movements, such as reaching or grasping. Usually this is achieved by targeting specific muscle groups which are associated to the targeted motion by experts. A general purpose, FES-based control theory capable of enabling neurological patients to achieve a wide range of positional goals in their peri-personal space is still missing. In this paper, we present an early analysis of the performance achievable through a muscular impedance control loop employing FES to actuate force and movement. The control is evaluated in a test where the user's upper limb is moved by means of an exonerve to a series of target positions on a plane without providing visual feedback nor requiring volitional effort. The results allow to characterize the performance of such a setup over time and to assess how well can it generalize over different target positions in the user's peri-personal space. The current study population also allows to evaluate the effects of user's experience with FES systems on the overall performance during the test. The results indicate that the proposed control loop can generalize well over different arm poses.

Trunk Posture from Randomly Oriented Accelerometers

Feedback control of functional neuromuscular stimulation has the potential to improve daily function for individuals with spinal cord injuries (SCIs) by enhancing seated stability. Our fully implanted networked neuroprosthesis (NNP) can provide real-time feedback signals for controlling the trunk through accelerometers embedded in modules distributed throughout the trunk. Typically, inertial sensors are aligned with the relevant body segment. However, NNP implanted modules are placed according to surgical constraints and their precise locations and orientations are generally unknown. We have developed a method for calibrating multiple randomly oriented accelerometers and fusing their signals into a measure of trunk orientation. Six accelerometers were externally attached in random orientations to the trunks of six individuals with SCI. Calibration with an optical motion capture system resulted in RMSE below 5° and correlation coefficients above 0.97. Calibration with a handheld goniometer resulted in RMSE of 7° and correlation coefficients above 0.93. Our method can obtain trunk orientation from a network of sensors without a priori knowledge of their relationships to the body anatomical axes. The results of this study will be invaluable in the design of feedback control systems for stabilizing the trunk of individuals with SCI in combination with the NNP implanted technology.

Improving the Learning Rate, Accuracy, and Workspace of Reinforcement Learning Controllers for a Musculoskeletal Model of the Human Arm

Cervical spinal cord injuries frequently cause paralysis of all four limbs - a medical condition known as tetraplegia. Functional electrical stimulation (FES), when combined with an appropriate controller, can be used to restore motor function by electrically stimulating the neuromuscular system. Previous works have demonstrated that reinforcement learning can be used to successfully train FES controllers. Here, we demonstrate that transfer learning and curriculum learning can be used to improve the learning rates, accuracies, and workspaces of FES controllers that are trained using reinforcement learning.

Feedback Control of Upright Seating with Functional Neuromuscular Stimulation during a Functional Task after Spinal Cord Injury: A Case Study

Seated stability is a major concern of individuals with trunk paralysis. Trunk paralysis is commonly caused by spinal cord injuries (SCI) at or above the thoracic spine. Current methods to improve stability restrict the movement of the user by constraining their trunk to an upright position. Feedback control of functional neuromuscular stimulation (FNS) can help maintain seated stability while still allowing the user to perform movements to accomplish functional tasks. In this study, an individual with a SCI (C7, AIS B) and an implanted stimulator capable of recruiting trunk and hip musculature unilaterally moved a weighted jar on a countertop to and from three prescribed stations directly in front, laterally, and across midline. For comparison, the tasks were performed with constant baseline stimulation and with feedback modulated stimulation based on the tilt of the trunk obtained from an external accelerometer fed into two PID controllers; one for forward trunk pitch and the other for lateral roll. The trunk pitch and roll angles were obtained through motion capture cameras and various measures of postural sway (95% fitted ellipse area, root mean squared (RMS), path length) and the repeatability (coefficient of variation (CoV), variance ratio (VR)) were calculated. Feedback control significantly increased RMS of trunk movement along the major axis of the fitted ellipse, but decreased RMS values during bending along the minor axis of motion. As a result, the fitted ellipse area decreased when deploying the jar to one of the stations and increased with the other two. The CoV indicated reduced variation in the presence of feedback controlled stimulation for all stations, and VR showed higher repeatability in trunk pitch. Plots of the trunk pitch and roll revealed a faster return to upright motion due to feedback stimulation.Clinical relevance- Feedback control in combination with FNS is a viable method to improve seated stability while still allowing dynamic movements in individuals with a SCI, thus addressing a major concern of the population.

Characterization of the Force Production Capabilities of Paralyzed Trunk Muscles Activated With Functional Neuromuscular Stimulation in Individuals With Spinal Cord Injury

Paralysis of the trunk results in seated instability leading to difficulties performing activities of daily living. Functional neuromuscular stimulation (FNS) combined with control systems have the potential to restore some dynamic functions of the trunk. However, design of multi-joint, multi-muscle control systems requires characterization of the stimulation-driven muscles responsible for movement.

OBJECTIVE

This study characterizes the input-output properties of paralyzed trunk muscles activated by FNS, and explores co-activation of muscles.

METHODS

Four participants with various spinal cord injuries (C7 AIS-B, T4 AIS-B, T5 AIS-A, C5 AIS-C) were constrained so lumbar forces were transmitted to a load cell while an implanted neuroprosthesis activated otherwise paralyzed hip and paraspinal muscles. Isometric force recruitment curves in the nominal seated position were generated by inputting the level of stimulation (pulse width modulation) while measuring the resulting muscle force. Two participants returned for a second experiment where muscles were co-activated to determine if their actions combined linearly.

RESULTS

Recruitment curves of most trunk and hip muscles fit sigmoid shaped curves with a regression coefficient above 0.75, and co-activation of the muscles combined linearly across the hip and lumbar joint. Subject specific perturbation plots showed one subject is capable of resisting up to a 300N perturbation anteriorly and 125N laterally; with some subjects falling considerably below these values.

CONCLUSION

Development of a trunk stability control system can use sigmoid recruitment dynamics and assume muscle forces combine linearly.

SIGNIFICANCE

This study informs future designs of multi-muscle, and multi-dimensional FNS systems to maintain seated posture and stability.

Hindsight Experience Replay Improves Reinforcement Learning for Control of a MIMO Musculoskeletal Model of the Human Arm

High-level spinal cord injuries often result in paralysis of all four limbs, leading to decreased patient independence and quality of life. Coordinated functional electrical stimulation (FES) of paralyzed muscles can be used to restore some motor function in the upper extremity. To coordinate functional movements, FES controllers should be developed to exploit the complex characteristics of human movement and produce the intended movement kinematics and/or kinetics. Here, we demonstrate the ability of a controller trained using reinforcement learning to generate desired movements of a horizontal planar musculoskeletal model of the human arm with 2 degrees of freedom and 6 actuators. The controller is given information about the kinematics of the arm, but not the internal state of the actuators. In particular, we demonstrate that a technique called "hindsight experience replay" can improve controller performance while also decreasing controller training time.

Predicting functional force production capabilities of upper extremity functional electrical stimulation neuroprostheses: a proof of concept study

OBJECTIVE

This study's goal was to demonstrate person-specific predictions of the force production capabilities of a paralyzed arm when actuated with a functional electrical stimulation (FES) neuroprosthesis. These predictions allow us to determine, for each hand position in a person's workspace, if FES activated muscles can produce enough force to hold the arm against gravity and other passive forces, the amount of force the arm can potentially exert on external objects, and in which directions FES can move the arm.

APPROACH

We computed force production predictions for a person with high tetraplegia and an FES neuroprosthesis used to activate muscles in her shoulder and arm. We developed Gaussian process regression models of the force produced at the end of the forearm when stimulating individual muscles at different wrist positions in the person's workspace. For any given wrist position, we predicted all possible forces a person can produce by any combination of individual muscles. Based on the force predictions, we determined if FES could produce force sufficient to overcome passive forces to hold a wrist position, the maximum force FES could produce in all directions, and the set of directions in which FES could move the arm. To estimate the error in our predictions, we then compared our force predictions based on single-muscle models to the actual forces produced when stimulating combinations of the person's muscles.

MAIN RESULTS

Our models classified the person's ability to hold static arm positions correctly for 83% (Session #1) and 69% (Session #2) for 39 wrist positions over two sessions. We predicted this person's ability to produce force at the end of her arm with an RMS error of 5.5 N and the percent of directions for which FES could achieve motion with RMS error of 10%. The accuracy of these predictions is similar to that found in the literature for FES systems with fewer degrees of freedom and fewer muscles.

SIGNIFICANCE

These person and device-specific predictions of functional capabilities of the arm allow neuroprosthesis developers to set achievable functional objectives for the systems they develop. These predictions can potentially serve as a screening tool for clinicians to use in planning neuroprosthetic interventions, greatly reducing the risk and uncertainty in such interventions.

Quantitative Assessment of Changes in Muscle Contractility Due to Fatigue During NMES: An Ultrasound Imaging Approach

OBJECTIVE

This paper investigates an ultrasound imaging-based non-invasive methodology to quantitatively assess changes in muscle contractility due to the fatigue induced by neuromuscular electrical stimulation (NMES).

METHODS

Knee extension experiments on human participants were conducted to record synchronized isometric knee force data and ultrasound images of the electrically stimulated quadriceps muscle. The data were first collected in a pre-fatigue stage and then in a post-fatigue stage. Ultrasound images were processed using a contraction rate adaptive speckle tracking algorithm. A two-dimensional strain measure field was constructed based on the muscle displacement tracking results to quantify muscle contractility.

RESULTS

Analysis of the strain images showed that, between the pre-fatigue and post-fatigue stages, there was a reduction in the strain peaks, a change in the strain peak distribution, and a decrease in an area occupied by the large positive strain.

CONCLUSION

The results indicate changes in muscle contractility due to the NMES-induced muscle fatigue.

SIGNIFICANCE

Ultrasound imaging with the proposed methodology is a promising tool for a direct NMES-induced fatigue assessment and facilitates new strategies to alleviate the effects of the NMES-induced fatigue.

Feedback control of functional electrical stimulation for arbitrary upper extremity movements

Functional electrical stimulation (FES) is a type of neuroprosthesis in which muscles are stimulated by electrical pulses in order to compensate for the loss of voluntary movement control. Modulating the stimulation intensities to reliably generate movements is a challenging control problem. For the first time, this paper presents a feedback controller for FES to control arm movements in a 2D (table-top) task space. This feedback controller is based on a recent human motor control model, which uses muscle synergies to simplify the calculations and improve control performance. The experimental results show that this control scheme can produce arbitrary movements in the 2D task space, with less than 2 cm hand position error from the specified targets.

A Modified Dynamic Surface Controller for Delayed Neuromuscular Electrical Stimulation

A widely accepted model of muscle force generation during neuromuscular electrical stimulation (NMES) is a second-order nonlinear musculoskeletal dynamics cascaded to a delayed first-order muscle activation dynamics. However, most nonlinear NMES control methods have either neglected the muscle activation dynamics or used an ad hoc strategies to tackle the muscle activation dynamics, which may not guarantee control stability. We hypothesized that a nonlinear control design that includes muscle activation dynamics can improve the control performance. In this paper, a dynamic surface control (DSC) approach was used to design a PID-based NMES controller that compensates for EMD in the activation dynamics. Because the muscle activation is unmeasurable, a model based estimator was used to estimate the muscle activation in realtime. The Lyapunov stability analysis confirmed that the newly developed controller achieves semi-global uniformly ultimately bounded (SGUUB) tracking for the musculoskeletal system. Experiments were performed on two able-bodied subjects and one spinal cord injury subject using a modified leg extension machine. These experiments illustrate the performance of the new controller and compare it to a previous PID-DC controller that did not consider muscle activation dynamics in the control design. These experiments support our hypothesis that a control design that includes muscle activation improves the NMES control performance.

Restoring standing capabilities with feedback control of functional neuromuscular stimulation following spinal cord injury

This paper reviews the field of feedback control for neuroprosthesis systems that restore advanced standing function to individuals with spinal cord injury. Investigations into closed-loop control of standing by functional neuromuscular stimulation (FNS) have spanned three decades. The ultimate goal for FNS standing control systems is to facilitate hands free standing and enabling the user to perform manual functions at self-selected leaning positions. However, most clinical systems for home usage currently only provide basic upright standing using preprogrammed stimulation patterns. To date, online modulation of stimulation to produce advanced standing functions such as balance against postural disturbances or the ability to assume leaning postures have been limited to simulation and laboratory investigations. While great technological advances have been made in biomechanical sensing and interfaces for neuromuscular stimulation, further progress is still required for finer motor control by FNS. Another major challenge is the development of sophisticated control schemes that produce the necessary postural adjustments, adapt against accelerating muscle fatigue, and consider volitional actions of the intact upper-body of the user. Model-based development for novel control schemes are proven and sensible approaches to prototype and test the basic operating efficacy of potentially complex and multi-faceted control systems. The major considerations for further innovation of such systems are summarized in this paper prior to describing the evolution of closed-loop FNS control of standing from previous works. Finally, necessary emerging technologies to for implementing FNS feedback control systems for standing are identified. These technological advancements include novel electrodes that more completely and selectively activate paralyzed musculature and implantable sensors and stimulation modules for flexible neuroprosthesis system deployment.

Semiparametric Identification of Human Arm Dynamics for Flexible Control of a Functional Electrical Stimulation Neuroprosthesis

We present a method to identify the dynamics of a human arm controlled by an implanted functional electrical stimulation neuroprosthesis. The method uses Gaussian process regression to predict shoulder and elbow torques given the shoulder and elbow joint positions and velocities and the electrical stimulation inputs to muscles. We compare the accuracy of torque predictions of nonparametric, semiparametric, and parametric model types. The most accurate of the three model types is a semiparametric Gaussian process model that combines the flexibility of a black box function approximator with the generalization power of a parameterized model. The semiparametric model predicted torques during stimulation of multiple muscles with errors less than 20% of the total muscle torque and passive torque needed to drive the arm. The identified model allows us to define an arbitrary reaching trajectory and approximately determine the muscle stimulations required to drive the arm along that trajectory.

Increases in human motoneuron excitability after cervical spinal cord injury depend on the level of injury

After human spinal cord injury (SCI), motoneuron recruitment and firing rate during voluntary and involuntary contractions may be altered by changes in motoneuron excitability. Our aim was to compare F waves in single thenar motor units paralyzed by cervical SCI to those in uninjured controls because at the single-unit level F waves primarily reflect the intrinsic properties of the motoneuron and its initial segment. With intraneural motor axon stimulation, F waves were evident in all 4 participants with C₄-level SCI, absent in 8 with C₅ or C₆ injury, and present in 6 of 12 Uninjured participants (P < 0.001). The percentage of units that generated F waves differed across groups (C₄: 30%, C₅ or C₆: 0%, Uninjured: 16%; P < 0.001). Mean (±SD) proximal axon conduction velocity was slower after C₄ SCI [64 ± 4 m/s (n = 6 units), Uninjured: 73 ± 8 m/s (n = 7 units); P = 0.037]. Mean distal axon conduction velocity differed by group [C₄: 40 ± 8 m/s (n = 20 units), C₅ or C₆: 49 ± 9 m/s (n = 28), Uninjured: 60 ± 7 m/s (n = 45); P < 0.001]. Motor unit properties (EMG amplitude, twitch force) only differed after SCI (P ≤ 0.004), not by injury level. Motor units with F waves had distal conduction velocities, M-wave amplitudes, and twitch forces that spanned the respective group range, indicating that units with heterogeneous properties produced F waves. Recording unitary F waves has shown that thenar motoneurons closer to the SCI (C₅ or C₆) have reduced excitability whereas those further away (C₄) have increased excitability, which may exacerbate muscle spasms. This difference in motoneuron excitability may be related to the extent of membrane depolarization following SCI.

NEW & NOTEWORTHY

Unitary F waves were common in paralyzed thenar muscles of people who had a chronic spinal cord injury (SCI) at the C₄ level compared with uninjured people, but F waves did not occur in people that had SCI at the C₅ or C₆ level. These results highlight that intrinsic motoneuron excitability depends, in part, on how close the motoneurons are to the site of the spinal injury, which could alter the generation and strength of voluntary and involuntary muscle contractions.

Bio-Inspired Controller on an FPGA Applied to Closed-Loop Diaphragmatic Stimulation

Cervical spinal cord injury can disrupt connections between the brain respiratory network and the respiratory muscles which can lead to partial or complete loss of ventilatory control and require ventilatory assistance. Unlike current open-loop technology, a closed-loop diaphragmatic pacing system could overcome the drawbacks of manual titration as well as respond to changing ventilation requirements. We present an original bio-inspired assistive technology for real-time ventilation assistance, implemented in a digital configurable Field Programmable Gate Array (FPGA). The bio-inspired controller, which is a spiking neural network (SNN) inspired by the medullary respiratory network, is as robust as a classic controller while having a flexible, low-power and low-cost hardware design. The system was simulated in MATLAB with FPGA-specific constraints and tested with a computational model of rat breathing; the model reproduced experimentally collected respiratory data in eupneic animals. The open-loop version of the bio-inspired controller was implemented on the FPGA. Electrical test bench characterizations confirmed the system functionality. Open and closed-loop paradigm simulations were simulated to test the FPGA system real-time behavior using the rat computational model. The closed-loop system monitors breathing and changes in respiratory demands to drive diaphragmatic stimulation. The simulated results inform future acute animal experiments and constitute the first step toward the development of a neuromorphic, adaptive, compact, low-power, implantable device. The bio-inspired hardware design optimizes the FPGA resource and time costs while harnessing the computational power of spike-based neuromorphic hardware. Its real-time feature makes it suitable for in vivo applications.

Brain-controlled muscle stimulation for the restoration of motor function

Loss of the ability to move, as a consequence of spinal cord injury or neuromuscular disorder, has devastating consequences for the paralyzed individual, and great economic consequences for society. Functional electrical stimulation (FES) offers one means to restore some mobility to these individuals, improving not only their autonomy, but potentially their general health and well-being as well. FES uses electrical stimulation to cause the paralyzed muscles to contract. Existing clinical systems require the stimulation to be preprogrammed, with the patient typically using residual voluntary movement of another body part to trigger and control the patterned stimulation. The rapid development of neural interfacing in the past decade offers the promise of dramatically improved control for these patients, potentially allowing continuous control of FES through signals recorded from motor cortex, as the patient attempts to control the paralyzed body part. While application of these 'brain-machine interfaces' (BMIs) has undergone dramatic development for control of computer cursors and even robotic limbs, their use as an interface for FES has been much more limited. In this review, we consider both FES and BMI technologies and discuss the prospect for combining the two to provide important new options for paralyzed individuals.

Further Results on Predictor-Based Control of Neuromuscular Electrical Stimulation

Electromechanical delay (EMD) and uncertain nonlinear muscle dynamics can cause destabilizing effects and performance loss during closed-loop control of neuromuscular electrical stimulation (NMES). Linear control methods for NMES often perform poorly due to these technical challenges. A new predictor-based closed-loop controller called proportional integral derivative controller with delay compensation (PID-DC) is presented in this paper. The PID-DC controller was designed to compensate for EMDs during NMES. Further, the robust controller can be implemented despite uncertainties or in the absence of model knowledge of the nonlinear musculoskeletal dynamics. Lyapunov stability analysis was used to synthesize the new controller. The effectiveness of the new controller was validated and compared with two recently developed nonlinear NMES controllers, through a series of closed-loop control experiments on four able-bodied human subjects. Experimental results depict statistically significant improved performance with PID-DC. The new controller is shown to be robust to variations in an estimated EMD value.

Multimodal decoding and congruent sensory information enhance reaching performance in subjects with cervical spinal cord injury

Cervical spinal cord injury (SCI) paralyzes muscles of the hand and arm, making it difficult to perform activities of daily living. Restoring the ability to reach can dramatically improve quality of life for people with cervical SCI. Any reaching system requires a user interface to decode parameters of an intended reach, such as trajectory and target. A challenge in developing such decoders is that often few physiological signals related to the intended reach remain under voluntary control, especially in patients with high cervical injuries. Furthermore, the decoding problem changes when the user is controlling the motion of their limb, as opposed to an external device. The purpose of this study was to investigate the benefits of combining disparate signal sources to control reach in people with a range of impairments, and to consider the effect of two feedback approaches. Subjects with cervical SCI performed robot-assisted reaching, controlling trajectories with either shoulder electromyograms (EMGs) or EMGs combined with gaze. We then evaluated how reaching performance was influenced by task-related sensory feedback, testing the EMG-only decoder in two conditions. The first involved moving the arm with the robot, providing congruent sensory feedback through their remaining sense of proprioception. In the second, the subjects moved the robot without the arm attached, as in applications that control external devices. We found that the multimodal-decoding algorithm worked well for all subjects, enabling them to perform straight, accurate reaches. The inclusion of gaze information, used to estimate target location, was especially important for the most impaired subjects. In the absence of gaze information, congruent sensory feedback improved performance. These results highlight the importance of proprioceptive feedback, and suggest that multi-modal decoders are likely to be most beneficial for highly impaired subjects and in tasks where such feedback is unavailable.