| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Department of Biomedical Engineering and Centre for Neuroscience, Faculty of Medicine and Dentistry, University of Alberta, Edmonton, Alberta, Canada
Submitted 5 November 2006; accepted in final form 10 January 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
). In 1911 Brown (1911) demonstrated that the spinal cord of adult cats is capable of generating rhythmic locomotor outputs in the absence of descending drive or sensory input. Later experiments supported the existence of locomotor-related neural circuitry within the spinal cord, now commonly referred to as the locomotor central pattern generator (CPG) (Grillner 1985; Orlovsky et al. 1999). Although the CPG is capable of generating a rhythmic output when activated by a tonic input in the form of electrical stimulation (Dimitrijevic et al. 1998; Guevremont et al. 2006; Magnuson and Trinder 1997; Shik 1997) or application of pharmacological agents (Cazalets et al. 1995; Cowley and Schmidt 1997; Kjaerulff and Kiehn 1996; Rossignol et al. 2001), it was shown to be modulated by afferent feedback. Limb position (Grillner and Rossignol 1978) and force (Duysens and Pearson 1980) are thought to play roles in the timing and duration of flexor (swing) and extensor (stance) activation during stepping (Norton et al. 2005). The goal of the current study was to evaluate the capacity of externally implemented controllers mimicking these physiological systems in restoring locomotion after neurological damage, such as a spinal cord injury.
Functional electrical stimulation (FES) is used as an interface with the nervous system to restore function to muscles and organs after disease or injury. The efficacy of the technique is based on the fact that after spinal cord injury the motoneurons below the level of the lesion remain intact and form a viable connection with the muscles they innervate. Electrical stimuli applied to intact nerves or muscles can then be used to generate functional contractions of targeted muscles. The clinical use of electrical stimulation dates back to the early 1960s when it was first used as a method for restoring lower (Liberson et al. 1961) and upper limb (Long and Masciarelli 1963) function after stroke or spinal cord injury. Currently available systems for restoring stepping require the user to initiate each step through the use of manual push buttons. In addition to being less desirable to the user, these hand-controlled systems have a higher metabolic cost than that of more automated controllers (Popovic et al. 2003). The ultimate system would be able to generate stable and reliable stepping in a manner similar to the way normal locomotion is controlled, using a combination of subconscious mechanisms with intermittent conscious effort.
Previous studies conducted in our laboratory demonstrated that intraspinal microstimulation can be used to generate in-place stepping in cats with spinal cord injury (Saigal et al. 2004). This was first achieved by applying predetermined phasic sequences of stimulation through groups of microwires generating reciprocal flexion and extension movements of the hindlimbs. Sensory-driven stepping was also achieved using a physiologically based if–then rule to govern the transitions between stance and swing (Prochazka 1996; Saigal et al. 2004).
In the present study we focused on investigating the capacity of two controllers to generate overground locomotion through the independent use of intrinsically timed (CPG) or sensory-driven phase transitions. Although it is unclear whether a CPG plays a significant role in human locomotion (Dimitrijevic et al. 1998), the intrinsic rhythmicity and ability to modulate output in response to external conditions makes the CPG an attractive model for the design of an FES control system. Our observations suggest that although both controllers were capable of generating overground locomotion, neither provided consistent load-bearing stepping. In addition, the intrinsically timed controller suffered from an inability to respond to changing internal and external conditions (i.e., fatigue and friction) and the sensory-driven controller exhibited high sensitivity to the selection of the initial parameters. For these reasons we suggest that a "combined controller," which uses sensory signals to modify an intrinsic rhythm may significantly improve the robustness of stepping. The combined controller may offer a more physiological representation of the neural control of overground locomotion by providing a balance between the roles of the CPG and sensory input. A controller was implemented and two experimental sessions were conducted to demonstrate the potential advantages of this combined approach.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Animal preparation and intramuscular electrode implantation
Anesthesia was temporarily induced through the inhalation of isoflurane (2–3% isofluorane in carbogen) and an intravenous catheter was inserted for administering sodium pentobarbital throughout the remainder of the experiment (0.25 mg/kg induction; maintenance 1:10 saline dilutions of the anesthetic). Sterilized IM electrodes were implanted bilaterally in six muscles that act to flex or extend the joints of the hindlimb. The implanted muscles were: sartorius anterior (hip flexor), semimembranosus anterior (hip extensor), biceps femoris posterior (knee flexor), vastus lateralis (knee extensor), tibialis anterior (ankle flexor), and gastrocnemius lateralis (ankle extensor). Intramuscular electrodes (nine-strand stainless steel Cooner wire, insulated except for 3- to 4-mm tips) were implanted near the motor point of the desired muscle according to the technique described by Basmajian and DeLuca (1985).
Experimental setup and data acquisition
After the implantation of IM wires the animal was transferred to a partial body support sling suspended from a sliding trolley (Fig. 1A). The trolley was placed on a track system allowing it to move with variable friction along the length of a 2.5-m custom-built walkway. The walkway consisted of two three-axis parallel force plates. The hindlimbs of the animal were placed on individual force plates allowing independent left and right force signals to be recorded along the entire length of the walkway during the stepping sequences. Accelerometers were secured firmly between the ankle and metatarsal phalangeal (MTP) joint of each hindlimb to provide a voltage signal representing the relative flexion or extension of the limb. Kinematic tracking markers were placed on the right side of the cat to indicate the iliac crest and the hip, knee, ankle, and MTP joints. Because of the movement of the skin over the knee joint during locomotion the knee joint marker could not be used to indicate the correct joint location. Therefore limb segment lengths were recorded and triangulation algorithms were used to provide a better estimate of the position of the knee joint.
|
Intramuscular stimulation protocol
Six dual-channel constant-current stimulators (EMS-6500, Electrostim Medical Services, Tampa, FL) were used to apply patterned trains of stimulation (biphasic charge balanced, 0.1–20 mA, 200-µs depolarizing pulse, 50 pulses/s) to activate flexor and extensor muscles in sequences that would generate coordinated stepping movements.
At the onset of each experiment the stimulation amplitude for motor threshold (level eliciting first visible muscle contraction) was determined as well as the stimulation amplitude required to generate the desired response. This was done using 1-s trains of stimulation applied to each individual muscle. After the stimulation amplitudes for individual muscles were obtained, synergies (groups of flexor or extensor muscles) were tested to generate functional movements of the whole limb with proportional changes in the joint angles throughout the action. Stimulation amplitudes were adjusted to evoke qualitatively balanced responses in the left and right legs. The adjusted stimulation amplitudes generating the desired muscle responses were recorded and used during the initial stepping trial as the plateau values for the stimulation envelopes. Further adjustments were made to these values from trial to trial based on observations made during the stepping sequences or by retesting the responses using 1-s-long stimulation trains applied to individual or synergistic muscles. This was done to achieve a combination of muscle activation levels that generated bilaterally balanced functional gait. This balance was subject to change as a result of factors such as the level of muscle fatigue and the amount of walkway friction at the onset of each stepping trial. This high dependency on initial parameter selection is consistent with results seen in two-dimensional (2D) (Yakovenko et al. 2004) and three-dimensional (3D) (Ekeberg and Pearson 2005) models of feline locomotion.
The control system
We designed three controllers that reflected some of the principles thought to govern locomotion in physiological systems: one that behaved like an intrinsically timed CPG, another that was a fully sensory-driven state controller, and a third that was a combination of intrinsically timed and sensory-driven control. Each controller used a different method to determine the timing and duration of the gait phases. However, a single output algorithm was used to translate these phases into stimulation waveforms based on predetermined stimulation levels. The controllers were tested during 14 recording sessions.
The intrinsically timed controller generated consistent patterns of flexion and extension based entirely on intrinsic timing in a manner similar to the output of an isolated CPG. The gait cycle of a cat can be divided into one flexion (F) and three extension (E1, E2, and E3) phases (Engberg and Lundberg 1969). A basic gait cycle consisting of a swing (hip, knee, and ankle flexion: F) phase and a stance (hip, knee, and ankle extension: E3) phase was implemented in all experiments. After preliminary tests using various phase durations, including those described by Halbertsma (1983), we used a 2-s gait cycle and set the stance and swing phase durations to 60 and 40%, respectively. Delayed retraction (hip extension) was added to this cycle so that the propulsive extension of the hip was initiated after the load-bearing extension of the knee and ankle was well established. In the final recording sessions we also added a short burst of stimulation to the knee flexor at the onset of swing to prevent toe drag (two sessions) and an increase in knee extensor stimulation during retraction to maintain ground contact (one session). All stimulation envelopes were trapezoidal in shape, consisting of three segments; 1) a ramp segment where the stimulation increased from motor threshold to the previously selected desired value, 2) a constant stimulation segment where the desired amplitude was maintained for the duration of the phase, and 3) a ramp segment where the stimulation amplitude returned from the desired level to motor threshold before the stimulation terminated. Step period, percentage of the phase spent in stance and swing, timing of the retraction phase, and duration of the ramp segments were set on the computer through a user interface to the control program. We did not experiment with different gait speeds using the intrinsically timed controller. However, this could be accomplished by changing the step period and adjusting the timing of the various gait phases as well as the stimulation amplitudes to achieve functional responses within a new phase duration. Figure 2A shows a conceptual illustration of a half-center CPG model that would generate continuous and rhythmic flexion and extension output patterns similar to those generated by our controller. Also shown is a schematic representation of stimulation envelopes achieved using the previously determined amplitudes with the intrinsically timed controller. Functional overground locomotion was first demonstrated in each cat using this controller before testing the other controllers. The activation profiles for the last step (to the right of the vertical dashed line, Fig. 2A) represent envelopes of electromyographic (EMG) data during overground locomotion in awake, intact cats, as collated from multiple sources by Yakovenko et al. (2002).
|
; Grillner and Rossignol 1978; McVea et al. 2005; Norton et al. 2005). The role of these signals during gait can be described by a set of if–then rules (Prochazka 1996) that we implemented using finite state control (Fig. 2B). The following rules were used to govern the swing-to-stance and stance-to-swing phase transitions: Stance-to-swing transitions:
IF the ipsilateral limb is extended
AND the ipsilateral limb is unloaded
AND the contralateral limb is loaded
THEN initiate swing in the ipsilateral limb
Swing-to-stance transitions:
IF the ipsilateral limb is flexed
THEN initiate stance in the ipsilateral limb
Although these rules allowed the left and right limbs to switch independently, they prevented the occurrence of a state of double swing and allowed a period of double stance during which both limbs were in the loaded phase of the gait cycle. The rule governing swing-to-stance allowed for early termination of swing if the stance leg became unloaded. This was done to avoid periods of no body support throughout the stepping sequences. Modulation of the stimulation amplitude applied to the knee extensor was also implemented such that if the phase transition rules were not met within a set period of time, the controller would increase the stimulation amplitude until either a maximum safe value was reached (20 mA) or the if–then rules were met. This feature enabled the animals to overcome walkway resistance in situations where the selected desired stimulation levels did not result in the generation of sufficient muscle force.
The states of limb extension/flexion and limb loading/unloading during stepping were determined using threshold-based measurements of signals from the accelerometers and force plates, respectively. Because of the movement of skin over the femur it is difficult to measure directly hip angle in cats. Therefore accelerometers secured to each foot of the animal were used to provide an estimate of the overall condition of flexion or extension of the limbs. The validity of this estimation is demonstrated by the examples of raw accelerometer signals and joint angles derived from video data shown in Fig. 1B. The joint angle data taken from FES evoked stepping can also be compared with the data shown in Fig. 1C, obtained during overground locomotion in an awake, intact cat. Thresholds were selected based on the ranges of limb position and GRF signals obtained during intrinsically timed CPG trials attained during the same recording session. The thresholds were typically set at 20% less than the maximal accelerometer signal for limb extension, 20% greater than the minimal accelerometer signal for limb flexion, and 50–75% of the average vertical force signals for limb loading. To reduce the number of false state transitions resulting from large movement-generated acceleration profiles during stepping (arrows in Fig. 2B), the limb position sensor signal was required to remain past the extension or flexion threshold for 100 ms before the state condition was met. Figure 2B uses sample feedback signals to illustrate the thresholds and resulting phase signals. Sensory-driven stepping often terminated in a state of double-unloaded extension in which both hindlimbs were extended behind the base of support of the cat.
A combined controller was implemented and tested in 2 recording sessions with the objective of preventing the occurrence of double-unloaded extension. This controller implemented the beneficial stability of the intrinsically timed system along with the state-dependent decision-making capacity of the sensory-driven system. The underlying switching was provided by an intrinsically timed CPG pattern that could be overridden based on two sensory rules whose purpose was to ensure limb loading at all times. The if–then rules for this controller were designed with the objective of preventing the occurrence of double-unloaded extension. The rules were:
GRF rule
IF the stance leg becomes unloaded
THEN terminate swing in the contralateral leg
Rolling rule
IF the forward progression stops
THEN take a step with each leg to a position under the body. Stand and push until the trolley starts to roll again.
An additional sensor (servo potentiometer) was added to the setup to provide a sawtooth displacement signal profile as the trolley rolled. When the unloaded condition of the GRF rule was met, the intrinsic timer was interrupted and reset to the point of termination of contralateral swing. When the rolling rule was invoked the intrinsic timer was paused in double stance and resumed only when the trolley movement was sensed.
All of the controllers were implemented in software (programmed in the high level computer language of C). The program was executed on a desktop computer interfaced to the stimulators through a National Instruments (NI) D/A card. Feedback signals from the accelerometers and force plates were digitized at 500 samples/s by an NI A/D card and smoothed by applying a 140-ms sliding window average.
Data analysis
Data analysis was completed using custom-written routines for Matlab V7.0 (The MathWorks, Natick, MA).
KINEMATICS. The video recordings were digitized and the x–y coordinates of each of the joint markers were extracted using a custom software package (MotionTracker2D) written in Matlab by Dr. Douglas Webber (University of Pittsburgh, Pittsburgh, PA). The motion of the right hindlimb was reconstructed using the coordinates of the markers as well as the measured lengths of the limb segments. The position of the knee joint was determined using triangulation methods because of the excessive movement of the skin over the joint during locomotion. Each frame was calibrated in the x- and y-axes to correct for skew introduced by the camera angle. The resulting values were used to reconstruct stick figures from which limb trajectories and joint excursions were extracted.
KINETICS. Vertical GRFs recorded during the evoked locomotion were used to quantify the functional value of stepping achieved. Force measurements were low-pass filtered (cutoff frequency 8 Hz) before analysis. Steps were divided into stance and swing phases based on output signals from the controller and mean vertical force was calculated for each individual step. Each stepping trial was initiated with a short period of double stance during which the animal typically lifted its hindquarters out of the support sling. The peak vertical force generated during this period of "forceful standing" was determined for each animal. As a result of the forceful nature of this movement this peak GRF value was assumed to be an overestimation of the weight of the hindquarters. Trials conducted in human volunteers performing sit-to-stand movements indicated that the weight of the individual (obtained during quiet standing) represents 85% of the peak force generated during the transition (data not shown). Therefore 85% of the peak vertical force recorded during the initial period of forceful standing in the animals was used as an approximation of the GRF required for full hindlimb loading.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
The locomotion achieved using a predetermined gait pattern generated by the intrinsically timed controller was characterized by rhythmic and consistent movements. Control parameters such as gait cycle duration, stance and swing phase durations, and stimulation amplitudes were selected at the onset of each trial and were invariable after the initiation of stepping. All stepping trials were conducted with a step cycle of 2 s with 60% (65% in one animal) of the duration spent in stance. Timing of the retraction phase was delayed from the initiation of stance by either 0 (hip extension occurs with knee and ankle extension), 200, 300, or 400 ms, or coincided with the initiation of contralateral stance (i.e., during double stance). The delay was selected to complement the forward and downward force generated by the knee extensor. Duration of the flexor burst at the onset of swing was typically 300–400 ms (37–50% of swing), allowing the toe to clear the surface while the leg was being swung forward by the ankle and hip flexors.
Figure 4 shows two 10-s clips taken from successful intrinsically timed trials in two different animals (refer to Fig. 1C for comparison with normal cat locomotion). The sequences of stick figures were reconstructed from kinematic data recorded during each of the full 10-s segments shown. Also shown are the calculated angles for the hip, knee, and ankle joints as well as filtered accelerometer and GRF signals. The phase determination signals in the bottom trace represent the stance (high) and swing (low) phase signals generated by the controller for the left (dotted) and right (solid) legs. The data displayed in Fig. 4A represent the final 10 s of a 17.5-s trial during which the animal took nine steps with each leg and traveled a total distance of 1.95 m. The maximal range of motion of the hip, knee, and ankle joints were 38, 35, and 50°, respectively, and represented 1.4-, 1.1-, and 1.4-fold of the ranges typically seen (28, 32, and 36°) in intact cats walking at slow to medium speeds (Rossignol et al. 1996). Left and right hindlimb vertical GRFs were 4.43 ± 1.25 and 5.48 ± 0.47 N (mean ± SD), respectively. These values represented roughly 10 and 12% of the total weight of the cat (3.68 kg) or 37 and 46% of the estimated force required for full hindlimb loading (11.9 N). The overlap between the force traces for the left and right legs indicates periods of double limb loading. This is a desired characteristic of functional stepping. This trial was terminated when the animal was no longer able to progress forward because of walkway friction and muscle fatigue. A characteristic of intrinsically timed control was that even after the forward progression stopped, the animal continued to step in-place (arrow) until the stimulation was terminated at the end of the trial. Figure 4B shows 10 s of data from the middle of an intrinsically timed trial that was deemed to be successful even though it was punctuated by slips of the right hindlimb. The total trial time was 27.5 s during which the animal took 13 steps with each leg and traveled a total distance of 2.5 m. Slips can be seen in the backward and upward trajectory of the MTP joint on the stick figures (as indicated by the direction of the arrow) and also as rapid increases in the limb angles at the end of the extension phase. The maximal range of motion of the hip, knee, and ankle joints was 55, 70, and 90°, respectively, and represented 1.8-, 2.2-, and 2.5-fold the typical joint ranges. Left and right hindlimb vertical GRFs were 8.51 ± 2.24 and 5.26 ± 1.10 N (mean ± SD), respectively. These values represented roughly 24 and 15% of the total weight of the cat (4.64 kg) or 67 and 41% of the estimated force required for full hindlimb loading (12.7 N). This trial was terminated when the animal reached the end of the walkway.
|
Sensory-driven state controller
We found that the stimulation amplitudes for successful intrinsically timed stepping had to be established to generate appropriate sensory-driven phase transitions. Each sensory-driven sequence was initiated with one intrinsically timed step. This placed the hindlimbs of the cat in an "acceptable" state of ipsilateral extension and unloading, and contralateral loading (if–then conditions required for stance-to-swing transitions) before the onset of sensory-driven switching. Poor parameter selection resulted in the "unacceptable" state of double-unloaded extension and an inability to switch to the swing phase of the gait cycle as a result of the unmet criteria of limb loading.
Figure 5 shows two examples of sensory-driven stepping obtained during a single recording session in one animal. The traces have the same format as those shown in Fig. 4 for intrinsically timed control with the exception of the phase determination signals, where the grayed regions during stance (high signal) represent periods during which the extensor stimulation amplitude was increased. This increase was triggered by a prolonged stance phase and resulted in an improvement in the ability of the animal to push the trolley in the presence of walkway resistance and muscle fatigue. The arrow at the bottom of the traces in A indicates the transition from intrinsically timed control (initiation steps) to sensory-driven control. The black dashed lines on the GRF and limb position traces indicate the threshold levels selected to represent limb extension (>2.4 V, 135°), flexion (<2.0 V, 100°), and loading (>4.4 N) for the right leg. Similar values were selected for the left leg (>2.7 V, <2.6 V, and >4.4 N, respectively) but are not illustrated here. The data shown in A represent a complete 25-s trial during which the animal took nine steps and traveled a total distance of 2.4 m. The initial step was taken using intrinsically timed parameters (2-s gait cycle, 60% stance, 6.3 ± 3.7 mA) with each of the following steps occurring when the if–then conditions for phase transitions were met. The cycle duration increased visibly as the animal progressed until it reached the end of the walkway and maintained double-loaded stance (standing). The stick figures illustrate only the last 10 s of these data starting at the point indicated by the arrow at the top of the traces. The sequence shown in B represents a trial that was terminated in double-unloaded extension. Note that the force profiles of the two legs decay simultaneously (circle in B), resulting in a state that does not meet any of the conditions for the initiation of swing. This is an unrecoverable and highly undesirable state in this system. This type of error was the most common mode of failure for sensory-driven stepping (Fig. 3) and represents a potentially dangerous situation in a clinical setting.
|
The difficulty in generating robust overground stepping using strictly intrinsically timed or sensory-driven control led us to design a controller that implemented both intrinsically timed and sensory-driven phase transitions. Two recording sessions were conducted using this controller to demonstrate the advantages of a combined design. The underlying phase switching was provided by an intrinsically timed pattern that could be overridden based on two sensory rules designed to prevent the occurrence of double-unloaded extension. A conceptual representation of this controller is illustrated in Fig. 6A. Figure 6B shows a 40-s data clip taken from a 90-s-long trial using the combined controller. The stick figures shown in B are reconstructed frame by frame from kinematic data. The threshold level indicated by the dashed line on the GRF trace shown in B was used by the GRF rule to terminate swing in the case where loading decayed in the stance leg. The rolling rule was implemented using the displacement signal (dashed) shown in the first row of traces. The displacement signal is shown in conjunction with the trigger signal (solid) that was generated by the controller to indicate that the trolley was rolling (low) or stationary (high). In this trial, the forward progression of the trolley was blocked by the experimenter (at the point indicated by the arrow in the stick figures) to test the functionality of the rolling rule. Note that, in B, after the trolley became stationary the legs took an additional step before halting in a double-loaded stance. This is a stable termination point. At this point assistance was provided in the form of a small push applied to the trolley for forward progression to be resumed. The controller responded to the movement of the trolley by reinitiating the intrinsically timed stimulation patterns. This resulted in the commencement of a new series of steps during which the GRF rule was evoked to correct for moments of unloading.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Controller structure and behavior
Three controllers were designed following some of the basic principles thought to govern locomotion in physiological systems. Two controllers were tested thoroughly over a series of 12 recording sessions: an intrinsically timed controller in which phase switching was based on predetermined patterns and a sensory-driven controller in which switching was based on feedback signals. A third controller, consisting of a combination of these two approaches, was implemented in two sessions to demonstrate the advantages of the combined control.
The controllers were designed using a hierarchical structure of three concurrently running processes that: 1) determined appropriate phase transitions, 2) formed stimulation output patterns, and 3) processed sensory inputs. In the intrinsically timed controller shown in Fig. 2A, the phase determination process generated rhythmic switching signals that were then used by the stimulation output process to generate appropriate stimulation waveforms for each muscle. This structure corresponds to a hierarchical model for the spinal control of locomotion consisting of separate rhythm generation and pattern formation layers. The hierarchical model was proposed based on observations of CPG-driven activity obtained during fictive locomotion in adult cats. In these conditions, the underlying rhythm was uninterrupted even in the presence of spontaneous deletions of flexor or extensor bursts in the respective nerves (Kriellaars et al. 1994; Lafreniere-Roula and McCrea 2005), suggesting that the input to motoneurons is generated independently from the oscillating circuitry. The presence of resetting and nonresetting deletions during fictive locomotion indicates that sensory signals can access both the rhythm generation and pattern formation layers of the CPG, respectively (Lafreniere-Roula and McCrea 2005).
This hierarchal structure was used in the present controllers because it allowed the timing of the gait phases (either through intrinsically timed or sensory-driven algorithms) to be controlled without knowledge of the actual stimulation profiles. Critical points in the gait cycle (such as the onset of stance, swing, and retraction) were dictated by the rhythm generation layer, which were then translated into appropriate stimulation profiles by the pattern formation layer.
Pattern formation
The patterns used at the onset of the study consisted of basic trapezoidal activation profiles with all extensors active during stance and all flexors during swing. Several modifications were subsequently made to the stimulation pattern produced by the controller to improve the quality of the generated gait. Comparisons of the stimulation envelopes produced by the controller to EMG patterns recorded in cats during locomotion (Yakovenko et al. 2002) indicated striking resemblances (i.e., increasing knee extensor activation during stance and bursting knee flexion at the onset of swing). Interestingly, this evolution toward "normal" activation profiles occurred entirely through efforts to improve the quality of gait with comparison to EMG data occurring only after all data had been obtained. This indicates the importance of appropriate activation profiles in the quality of overground stepping achieved. It also illustrates that it is possible to generate overground stepping through the activation of only a subset of the many muscles normally active during locomotion.
Rhythm generation and sensory integration
The behavior of the intrinsically timed controller was similar to that of a reciprocally coupled half-center as proposed by Brown (1911, 1914) and illustrated in Fig. 2A. The controller generated a periodic pattern of flexor and extensor activity in the absence of input signals. The sensory-driven controller more closely resembled the theory proposed by Sherrington (1910) in which stepping is generated solely through a chain of reflexes. The controller had no intrinsic rhythmicity and the switching between flexor and extensor activity was completely dependent on sensory feedback signals. The feedback signals were also used by the controller to modulate the stimulation pattern (e.g., increase stimulation to the knee extensor when a prolonged stance phase was detected).
Signals representing limb position and GRF were selected for the sensory-driven controller because of their roles in timing gait transitions in insects (Bassler and Buschges 1998; Pearson and Duysens 1976), cats (Grillner and Rossignol 1978; McVea et al. 2005; Norton et al. 2005; Pearson 1995; Whelan et al. 1995), and human infants (Pang and Yang 2000). Experiments conducted in cats indicated that information about hip angle and limb loading is derived from muscle spindles in hip flexors (Hiebert et al. 1996) and Golgi tendon organs in ankle extensors (Pearson et al. 1998; Whelan et al. 1995), respectively. Both hip angle and GRF play important roles in regulating stepping, as shown in cats with complete spinal cord injury and selective deafferentations (Norton et al. 2005). Cutaneous sensory input also has a modulatory effect on the gait pattern (Bouyer and Rossignol 2003a,b). This input represents loading information that is provided through pathways other than those involving the Golgi tendon organs. To simplify the design of our controllers we removed the redundancy of the cutaneous input that is normally present in natural systems.
In the present study, the sensory signals were interfaced with both the rhythm generation (for phase transitions) and the pattern formation (for stimulation amplitude modulation) layers of the controller. Feedback to the pattern formation layer acted to reinforce the pattern rather than generate deletions as seen during fictive locomotion. Deletion of an extensor phase without a compensatory action would result in a loss of loading in our animal and was therefore not desirable.
Although neither the intrinsically timed nor the sensory-driven phase transitions consistently produced functional locomotion, we found that the intrinsically timed stepping was typically more robust than the sensory-driven stepping. This reinforces the idea that sensory information may actually decrease the stability of a system by adding signal noise (Kuo 2002) as well as requiring additional control parameters. However, the intrinsically timed controller was unable to respond to changes in the state of stepping caused by loss of traction, variable walkway resistance, or muscle fatigue. For these reasons we proposed the use of a third controller consisting of an underlying rhythm that could be reset or modified as indicated by sensory feedback signals.
The combined controller (Fig. 6A) was constructed from the phase determination and pattern formation stages of the intrinsically timed controller, with the addition of a sensory input stage thay acted to modify the switching of the flexion and extension half-centers. Two recording sessions were conducted to demonstrate the concept of a combined controller and the results indicate its potential in providing a versatile foundation for generating robust stepping. The sensory rules used in this session were designed specifically to prevent periods of insignificant limb loading by influencing the rhythm generation layer.
The design of a combined controller is not limited to the two sensory rules implemented in this study and its performance could potentially be improved by designing additional sensory rules that also modulate the pattern formation layer. The rules implemented in this study were selected to prevent the critical modes of failure seen in our model. Rules specific to bipedal gait could be selected and implemented clinically. These rules could provide features such as stimulation amplitude modulation in the presence of muscle fatigue and phase duration modulation either to induce or to respond to changes in forward velocity. Buttons or switches allowing direct user intervention can also be implemented with the combined controller to address situations where more specific control is required (such as going up or down stairs). These mechanisms would reflect the natural ability to provide conscious control of gait in challenging conditions.
Parameter selection
We found that the proper selection of stimulation parameters was critical for achieving successful stepping sequences. Despite the thought that feedback control should allow the system to adjust to small imbalances, parameter selection was particularly critical when using the sensory-driven controller. This could be explained by the fact that the sensory signals were used strictly to generate phase transitions. In physiological systems it is thought that reinforcing/inhibitory reflex pathways play a role in adjusting the muscle activation levels in response to the specific needs at any point throughout the gait cycle (Pearson and Duysens 1976). These reflexes act to increase the force output during stance and limit the amplitude of the movement during swing. Implementing such feedback loops into our controller may have widened the range of acceptable starting parameters for successful stepping sequences. However, they would have also had the effect of introducing additional parameters (such as acceptable sensor signal ranges and thresholds) that may have ultimately required additional tuning for functional operation. In addition, the use of IM stimulation of a select number of muscles limits the spectrum of movements that can be generated. The differences in recruitment characteristics resulting from muscle properties and IM electrode placement make synergistic movements particularly difficult to generate because of the requirement for the balanced contributions of various muscles over a range of stimulation levels. The combined controller had the benefit of using a simple interpretation of sensor signals to produce only the necessary changes to an underlying rhythm.
Control algorithms using hip angle and GRF feedback signals were also previously implemented in 2D (Yakovenko et al. 2004) and 3D (Ekeberg and Pearson 2005) computer models of cat hindlimb locomotion. Results from the 2D model indicated that the use of state-dependent control rules similar to the ones implemented in our sensory-driven controller increased the range of parameters suitable for stable locomotion and enabled the model to step at different velocities. The fact that our sensory-driven controller suffered from reduced stability is likely attributable to the increased complexity of the in vivo system. Additional variability was introduced by sensor noise, as well as by muscle fatigue and recruitment properties. Unlike the computer simulations, two consecutive trials performed in our in vivo system would not generate identical results. For this reason we did not perform a detailed analysis of the stable parameter space in our system.
The 3D cat model developed by Ekeberg and Pearson (2005) was used to examine the relative roles of hip angle and GRF information in the appropriate initiation of stance-to-swing phase transitions during overground locomotion. They implemented a state controller similar to the one used in our sensory-driven system. Initially, hip angle and GRF were combined linearly to determine phase transitions (i.e., low force would initiate flexion at less extension, whereas high force would require more extension). They then removed the contribution of hip angle or GRF from the controller and observed the effect on the quality of stepping achieved. Coordinated stepping could be generated when GRF was used alone but hip angle alone resulted in decoupled gait. This supports the idea that the point in the gait cycle where weight is transferred from one limb to the other is critical and must be carefully regulated, as is done through the use of our if–then rules. Even though both models (Ekeberg and Pearson 2005; Yakovenko et al. 2004) were able to generate stable locomotion, they both suffered from high sensitivity to parameter selection. However, a simple pendulum model of rhythmic limb movements demonstrated that a combination of feedforward (intrinsically timed) and feedback (sensory-driven) control can provide improved stability (Kuo 2002).
Although the goal of our study was not to reproduce the aforementioned models, our findings in an in vivo system extend some of the results achieved in the computer simulations by incorporating real muscle properties. We found that in the in vivo system, the balance of muscle activation through FES becomes increasingly difficult to achieve as a result of the nonlinear responses to stimulation during individual movements (arising from the order of muscle fiber recruitment and hysteresis) and between consecutive steps and trials (arising from muscle fatigue). This is a property that is not necessarily present in computer models, where the same set of initial conditions can be replicated and parameters tuned using optimizing algorithms. The use of stimulation applied at 50 pulses/s may also add to the instability of the responses resulting from the extended plateau region of the length–tension relationship of near-tetanized muscle and the sharp decay in force that occurs at short muscle lengths. The generation of locomotion in an intact physiological system is a complex task in which threshold levels and feedback signals are used dynamically to regulate phase transitions and muscle activation levels. Our efforts in achieving stable locomotion using strictly intrinsically timed or sensory-driven phase transitions leads us to propose that a combined controller, consisting of a balance of these two components, may provide the most robust locomotion while still providing a fair representation of the physiological control of locomotion. As a corollary, the presence of "threshold levels" of supraspinal and sensory inputs in individuals with incomplete spinal cord injury was recently suggested as a necessary factor for translating the benefits of body weight–supported treadmill locomotor training to functional gains in overground ambulation (Dobkin et al. 2006).
Clinical implications
Electrical stimulation to restore movement can be controlled through either open- or closed-loop control. Comparatively simple systems that aid in restoring gait such as foot-drop stimulators use closed-loop control with stimulation being driven by signals such as leg angle (Dai et al. 1996) or foot contact (Burridge et al. 1997) derived automatically during the step cycle. More complicated systems that require several channels of stimulation to restore stepping in users with complete motor paralysis typically operate under open-loop control in which users initiate each step, or part of a step, through manual push buttons.
The primary benefit of the controllers implemented in this study is that each is able to produce functional outputs without the conscious intervention of the user. The output patterns are either generated in a predetermined manner, driven by feedback signals as the movements progressed, or preferably by a combination of the two.
Other groups developed control strategies for restoring walking after spinal cord injury. A dual-level (pattern generator/pattern shaper) adaptive feedforward controller that can track cyclic knee torques during knee flexion and extension movements while sitting was developed (Abbas and Chizeck 1995; Abbas and Triolo 1997). This controller used cycle-to-cycle corrections to make appropriate modifications to the stimulation applied during the subsequent cycles. The adaptive feedforward system was expected to compensate for slow changes in muscle properties and the addition of a feedback loop would allow the controller to respond rapidly to mechanical perturbations. Reiss and Abbas (2001) demonstrated that angle feedback signals can be used by the feedforward controller to track cyclic movements during muscle fatigue. The sensitivity of our system to the selection of stimulation amplitudes suggests that the addition of cycle-to-cycle feedforward correction may improve the gait by changing the stimulation levels to maintain consistent movements over multiple steps.
If–then rule-based state controllers have been used in other systems to restore mobility. In one system, signals from goniometers, accelerometers, and load sensors on crutches were used to determine the users' current position (seated or standing), their intention to move (stand or step), and make discrete state transitions throughout the movement (Veltnik et al. 1996). A feedback controller that used if–then sensory rules determined from machine learning algorithms was implemented in a subject with complete paraplegia and yielded good overground stepping (Fisekovic and Popovic 2001). In contrast to these systems, our sensory-driven controller is based on physiological if–then rules for locomotion. We started with a fundamental set of if–then rules and a simple output activation pattern and added features as required to improve the quality of stepping achieved. By taking this bottom-up approach we were able to examine the roles of the control rules and activation patterns in generating stable gait.
Our sensory-driven controller uses limb position and GRF feedback signals with if–then rules to generate appropriate phase transitions. Accelerometers have already been used on humans to measure effectively knee joint angles during standing (Veltink and Franken 1996) or tilt of the shank during stepping (Dai et al. 1996). These results suggest that the direct measurement of hip angle using accelerometers secured to the thigh will be feasible in a clinical setting. We used force plates to obtain GRF information because they generate signals that are simple to interpret and have a large signal-to-noise ratio. For clinical implementation, force-sensing resistors (FSRs) placed in the sole of the shoe can be used to provide information about the distribution of pressure under the foot. These sensors have been used to detect events in the gait cycle during walking (Skelly and Chizeck 2001). Pressure-related sensory signals can also be obtained through an implanted cuff placed on the sural nerve to record electroneurographic (ENG) signals carried by cutaneous sensory nerve fibers originating from the lateral sole of the foot. It was previously shown in cats that small ENG signals (5–10 µA) can be recorded in the presence of stimulation evoked muscle activity (10–100 mV) by using a tripolar recording cuff and simple data processing methods (Haugland and Hoffer 1994). These nerve signals can then be used by a state controller to determine gait events during FES walking (Strange and Hoffer 1999). Adaptive logic networks (ALNs) can be used to interpret these signals even when walking on various surfaces or while wearing different footwear (Hansen et al. 2004).
A study conducted to evaluate four different modes of FES stepping (hand control and automatic control of slow walking, near-normal walking, and near-ballistic walking) indicated that the metabolic cost and physiological cost index of hand-controlled phase transitions were higher compared with those of more automated forms of control (Popovic et al. 2003). Although ballistic walking was shown to have the lowest cost, five of the six subjects selected the automatic control of slow walking to be their preferred mode because of the need to coordinate upper and lower limb movements during stepping. The control system was designed using a 3D model of walking that implemented neural networks to generate predetermined stimulation patterns that activated muscles in sequences corresponding to the desired gait. A system such as this one could be used in conjunction with our combined controller to generate appropriate intrinsically timed muscle activation sequences (as determined by the model and neural network), which could then be modified by sensory signals. These feedback signals could improve the ability of the user to coordinate upper and lower limb movements by ensuring that the stepping remains within the individual's base of support. Push buttons could also be provided, allowing the user to intervene actively in cases where the intrinsic pattern is not suitable for the desired task. The ultimate combined control system would achieve a balance between the low metabolic costs of a fully automated system and the increased biomechanical stability experienced by users of manually controlled systems. The use of a more fatigue-resistant stimulation paradigm such as intraspinal microstimulation might result in further improvements to the quality of stepping achieved (Saigal et al. 2004).
In conclusion, in this study we have demonstrated that although FES-evoked overground locomotion can be generated using either intrinsically timed or sensory-driven controllers, the stepping achieved lacks the robustness and load-bearing qualities characteristic of functional locomotion. We demonstrated that intrinsically timed stepping is less susceptible to failure as a result of poor initial parameter selection but that sensory-driven stepping has the advantage of being able to adapt to changes in walkway friction or muscle properties during the stepping sequence. A combined controller that has an underlying rhythmic output that can be modified by sensory signals would best resemble the rhythmic output of a locomotor CPG in a physiological system. A combined control system such as this one may provide a means for restoring robust and functional overground locomotion for people with spinal cord injury.
|
GRANTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: V. K. Mushahwar, 513 Heritage Medical Research Centre, University of Alberta, Edmonton, Alberta T6G 2S2, Canada (E-mail: vivian.mushahwar{at}ualberta.ca)
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Abbas JJ, Triolo RJ. Experimental evaluation of an adaptive feedforward controller for use in functional neuromuscular stimulation systems. IEEE Trans Rehabil Eng 5: 12–22, 1997.[CrossRef][Medline]
Basmajian JV, DeLuca CJ. Muscles Alive. Their Functions Revealed by Electromyography. Baltimore, MD: Williams & Wilkins, 1985.
Bassler U, Buschges A. Pattern generation for stick insect walking movements—multisensory control of a locomotor program. Brain Res Rev 27: 65–88, 1998.[CrossRef][Medline]
Bouyer LJ, Rossignol S. Contribution of cutaneous inputs from the hindpaw to the control of locomotion. I. Intact cats. J Neurophysiol 90: 3625–3639, 2003a.
Bouyer LJ, Rossignol S. Contribution of cutaneous inputs from the hindpaw to the control of locomotion. II. Spinal cats. J Neurophysiol 90: 3640–3653, 2003b.
Brown TG. The intrinsic factors in the act of progression in the mammal. Proc R Soc Lond B Biol Sci 84: 309–318, 1911.
Brown TG. On the nature of the fundamental activity of the nervous centers; together with an analysis of the conditioning of rhythmic activity in progression, and a theory of the evolution of function in the nervous system. J Physiol 48: 18–46, 1914.
Burridge JH, Taylor DN, Hagan SA, Wood DE, Swain ID. The effects of common peroneal stimulation on the effort and speed of walking: a randomized controlled trial with chronic hemiplegic patients. Clin Rehabil 11: 201–210, 1997.
Cazalets JR, Borde M, Clarac F. Localization and organization of the central pattern generator for hindlimb locomotion in newborn rat. J Neurosci 15: 4943–4951, 1995.[Abstract]
Cowley KC, Schmidt BJ. Regional distribution of the locomotor pattern-generating network in the neonatal rat spinal cord. J Neurophysiol 77: 247–259, 1997.
Dai R, Stein RB, Andrews BJ, James KB, Wieler M. Application of tilt sensors in functional electrical stimulation. IEEE Trans Rehabil Eng 4: 63–72, 1996.[CrossRef][Medline]
Dimitrijevic MR, Gerasimenko Y, Pinter MM. Evidence for a spinal central pattern generator in humans. Ann NY Acad Sci 860: 360–376, 1998.[CrossRef][Web of Science][Medline]
Dobkin B, Apple D, Barbeau H, Basso M, Behrman A, Deforge D, Ditunno J, Dudley G, Elashoff R, Fugate L, Harkema S, Saulino M, Scott M. Weight-supported treadmill vs over-ground training for walking after acute incomplete SCI. Neurology 66: 484–493, 2006.
Duysens J, Pearson KG. Inhibition of flexor burst generation by loading ankle extensor muscles in walking cats. Brain Res 187: 321–332, 1980.[CrossRef][Web of Science][Medline]
Ekeberg O, Pearson K. Computer simulation of stepping in the hind legs of the cat: an examination of mechanisms regulating the stance-to-swing transition. J Neurophysiol 94: 4256–4268, 2005.
Engberg I, Lundberg A. An electromyographic analysis of muscular activity in the hindlimb of the cat during unrestrained locomotion. Acta Physiol Scand 75: 614–630, 1969.[Web of Science][Medline]
Fisekovic N, Popovic DB. New controller for functional electrical stimulation systems. Med Eng Phys 23: 391–399, 2001.[CrossRef][Web of Science][Medline]
Grillner S. Neurobiological bases of rhythmic motor acts in vertebrates. Science 228: 143–149, 1985.
Grillner S, Rossignol S. On the initiation of the swing phase of locomotion in chronic spinal cats. Brain Res 146: 269–277, 1978.[CrossRef][Web of Science][Medline]
Guevremont L, Renzi CG, Norton JA, Kowalczewski J, Saigal R, Mushahwar VK. Locomotor-related networks in the lumbosacral enlargement of the adult spinal cat: activation through intraspinal microstimulation. IEEE Trans Neural Syst Rehabil Eng 14: 266–272, 2006.[CrossRef][Web of Science][Medline]
Halbertsma JM. The stride cycle of the cat: the modelling of locomotion by computerized analysis of automatic recordings. Acta Physiol Scand Suppl 521: 1–75, 1983.[Medline]
Hansen M, Haugland MK, Sinkjaer T. Evaluating robustness of gait event detection based on machine learning and natural sensors. IEEE Trans Neural Syst Rehabil Eng 12: 81–88, 2004.[CrossRef][Web of Science][Medline]
Haugland MK, Hoffer JA. Artifact-free sensory nerve signals obtained from cuff electrodes during functional electrical stimulation of nearby muscles. IEEE Trans Neural Syst Rehabil Eng 2: 37–40, 1994.
Hiebert GW, Whelan P, Prochazka A, Pearson KG. Contribution of hind limb flexor muscle afferents to the timing of phase transitions in the cat step cycle. J Neurophysiol 75: 1126–1137, 1996.
Kjaerulff O, Kiehn O. Distribution of networks generating and coordinating locomotor activity in the neonatal rat spinal cord in vitro: a lesion study. J Neurosci 16: 5777–5794, 1996.
Kriellaars DJ, Brownstone RM, Noga BR, Jordan LM. Mechanical entrainment of fictive locomotion in the decerebrate cat. J Neurophysiol 71: 2074–2086, 1994.
Kuo AD. The relative roles of feedforward and feedback in the control of rhythmic movements. Motor Control 6: 129–145, 2002.[Web of Science][Medline]
Lafreniere-Roula M, McCrea DA. Deletions of rhythmic motoneuron activity during fictive locomotion and scratch provide clues to the organization of the mammalian central pattern generator. J Neurophysiol 94: 1120–1132, 2005.
Liberson WT, Holmquest HJ, Scot D, Dow M.Functional electrotherapy: stimulation of the peroneal nerve synchronized with the swing phase of the gait of hemiplegic patients. Arch Phys Med Rehabil 42: 101–105, 1961.[Medline]
Long C, Masciarelli V. An electrophysiological splint for the hand. Arch Phys Med Rehabil 44: 449–503, 1963.
Magnuson DSK, Trinder TC. Locomotor rhythm evoked by ventrolateral funiculus stimulation in the neonatal rat spinal cord in vitro. J Neurophysiol 77: 200–206, 1997.
McVea DA, Donelan JM, Tachibana A, Pearson KG. A role for hip position in initiating the swing-to-stance transition in walking cats. J Neurophysiol 94: 3497–3508, 2005.
Norton J, Guevremont L, Mushahwar V. Significance of sensory input for stepping in chronically spinalized cats. Soc Neurosci, 630.5, 2005.
Orlovsky GN, Deliagina TG, Grillner S. Neuronal Control of Locomotion: From Mollusc to Man. New York: Oxford Univ. Press, 1999.
Pang MY, Yang JF. The initiation of the swing phase in human infant stepping: importance of hip position and leg loading. J Physiol 528: 389–404, 2000.
Pearson KG. Proprioceptive regulation of locomotion. Curr Opin Neurobiol 5: 786–791, 1995.[CrossRef][Web of Science][Medline]
Pearson KG, Duysens J. Function of segmental reflexes in control of stepping in cockroaches and cats. In: Neural Control of Locomotion, edited by Herman RM, Grillner S, Stein PSG, Stuart DG. New York: Plenum Press, 1976, p. 519–538.
Pearson KG, Misiaszek JE, Fouad K. Enhancement and resetting of locomotor activity by muscle afferents. Ann NY Acad Sci 860: 203–215, 1998.[CrossRef][Web of Science][Medline]
Popovic D, Radulovic M, Schwirtlich L, Jaukovic N. Automatic vs hand-controlled walking of paraplegics. Med Eng Phys 25: 63–73, 2003.[CrossRef][Web of Science][Medline]
Prochazka A. Proprioceptive feedback and movement regulation. In: Handbook of Physiology. Exercise: Regulation and Integration of Multiple Systems. Neural Control of Movement. Bethesda, MD: Am. Physiol. Soc., 1996, sect. 12, p. 89–127.
Riess J, Abbas JJ. Adaptive control of cyclic movements as muscles fatigue using functional neuromuscular stimulation. IEEE Trans Neural Syst Rehabil Eng 9: 326–330, 2001.[CrossRef][Web of Science][Medline]
Rossignol S, Giroux N, Chau C, Marcoux J, Brustein E, Reader TA. Pharmacological aids to locomotor training after spinal injury in the cat. J Physiol 533: 65–74, 2001.
Saigal R, Renzi C, Mushahwar VK. Intraspinal microstimulation generates functional movements after spinal-cord injury. IEEE Trans Neural Syst Rehabil Eng 12: 430–440, 2004.[CrossRef][Web of Science][Medline]
Sherrington CS. Flexion-reflex of the limb, crossed extension-reflex, and reflex stepping and standing. J Physiol 40: 28–121, 1910.
Shik ML. Recognizing propriospinal and reticulospinal systems of initiation of stepping. Motor Control 1: 310–313, 1997.
Skelly MM, Chizeck HJ. Real-time gait event detection for paraplegic FES walking. IEEE Trans Neural Syst Rehabil Eng 9: 59–68, 2001.[CrossRef][Web of Science][Medline]
Strange KD, Hoffer JA. Restoration of use of paralyzed limb muscles using sensory nerve signals for state control of FES-assisted walking. IEEE Trans Rehabil Eng 7: 289–300, 1999.[CrossRef][Medline]
Veltnik PH, de Vries W, Hermans HJ, Baardman G, Ijzerman M, Heinze S, Nene AV, Zilvold G, Boom HBK. A comprehensive FES control system for mobility restoration in paraplegics. In: Neuroprosthetics from Basic Research to Clinical Applications, edited by Pedotti A, Ferrarin M, Quintern J, Riener R. New York: Springer-Verlag, 1996, p. 163–168.
Veltink PH, Franken HM. Detection of knee unlock during stance by accelerometry. IEEE Trans Rehabil Eng 4: 395–402, 1996.[CrossRef][Medline]
Whelan PJ, Hiebert GW, Pearson KG. Stimulation of the group I extensor afferents prolongs the stance phase in walking cats. Exp Brain Res 103: 20–30, 1995.[Web of Science][Medline]
Yakovenko S, Gritsenko V, Prochazka A. Contribution of stretch reflexes to locomotor control: a modeling study. Biol Cybern 90: 146–155, 2004.[CrossRef][Web of Science][Medline]
Yakovenko S, Mushahwar V, VanderHorst V, Holstege G, Prochazka A. Spatiotemporal activation of lumbosacral motoneurons in the locomotor step cycle. J Neurophysiol 87: 1542–1553, 2002.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
